How do I narrate a player's PC's actions without causing unintended consequences for the PC?

As DM, I wouldn’t dictate the details of the investigation of statue 1, as you do. Ideally, I would ask them to describe how they are investigating—before rolling. To keep things moving, I have been known to allow a roll and reveal the results before the description, since it’s easier to describe what you were doing when you know how it turns out, but knowing the trap was coming up, I probably wouldn’t have.

But if, for the sake of argument, I did that shortcut on statue 1, on statue 2 I would ask them if they were taking the same approach with statue 2 when the player announced they were moving on. Keeping it casual can avoid giving too much away: “Let’s see what’s going on with statue 2,” “Same approach?” With paranoid players, they might catch that something is up, but oh well: I trust my players. Either they’ll resist the urge to metagame, or they won’t and I’ll just chalk that one up to them deciding they’d have more fun that way. The game is there for having fun, after all; metagaming is bad for the game, but not necessarily always the worst thing for the game.

System agnostic answer

Many roleplay games (D&D traditionally not among them) have the concept of "failing forward". This means that every roll has some consequence which is usually narrated by the GM and usually bad for the character or the party as a whole. A roll without possible negative consequences should not be made.

Examples of such rules are:

Every moment of play, roll dice or say “yes.”

If nothing is at stake, say “yes” [to the player’s request], whatever they’re doing. Just go along with them. If they ask for information, give it to them. If they have their characters go somewhere, they’re there. If they want it, it’s theirs.

Sooner or later—sooner, because [your game’s] pregnant with crisis— they’ ll have their characters do something that someone else won’ t like. Bang! Something’s at stake. Start the confl ict and roll the dice.

Roll dice, or say “yes.” (Dogs in the Vineyard)

Burning Wheel has "Intent" and "Let it ride" rules. Intent is stated before every roll. When the roll succeeds, the intent happens. When the roll fails, all manners of things may happen (at GMs discretion) but the intent definitively doesn't happen. This can't be changed by any roll until the situation in fiction has significantly changed. An example of this two rules in play would be:

Player: I want to pick the lock before the guards come around.

roll fail

GM: Yeah, you open the lock... just as a guard comes around the corner.

Apocalypse World has "moves" a GM (MC in Apocalypse World) can make when players fail a roll. The moves are of general nature ("take their stuff away", "separate them", "inflict harm") and don't have to be directly caused by the attempted action.

What that means for your situation

With the statue example this would mean that a failed roll would totally allow you to narrate the character springing the trap. He chose to investigate and thus took the risk of touching something, he shouldn't touch. If he insists on doing a "safe, no touch" investigation he most likely won't get much useful information (but also wouldn't have to roll).

Final advice

While some games support, encourage and even demand to be played like this, many games (including D&D) don't. Thus any attempt to start playing in this style has to be agreed upon with the whole group (trivially true when it is part of the official rules of the game being played). I strongly suggest shifting to "failing forward" as the alternative leads to very slow and static play-style where not much happens, nobody trusts one another and where there is constant rules lawyering.

When the roll results can give hints to the player I advise to have the DM make them in secret. In this way the DM prevents metagaming based on the result.

Coming to the answer: the DM knows the second statue has a contact trap so he should already be thinking from the beginning about the problem, and ask the player to describe how exactly he investigates the first statue, and then the second one. If the player doesn't say anything about touching or not touching (can happen) the DM can just ask. The player will be suspicius but he doesn't know which will be good or bad. Also, the DM can start the description assuming the PC doesn't touch, give some information, and then ask if he wants investigate more closely by touching the statue. The DM should do this even when there isn't a trap! Keep them alert!

About the example in the comments:

For things like that, the DM can assume what he wants. He pays, he gives money, mechanically doesn't make difference how he gives the money... the rules don't go so far. If the player wants to do something strange he should say so.

When the player doesn't have any hint the DM also can roll a die for a random solution.

"I'd like to roll an investigate check on Statue 1"

That's not really what a player should say, at least in games with a traditional role split between players (state their characters' actions) and a GM/DM (adjudicates those actions). Since "DM" is used in the question, I'll assume such a traditional-role game.

In such a game, the player shouldn't really "interact with the game using rules." Instead, they should "interact with the world using actions." In other words, the player describes character actions and based on this, the DM chooses which rules, if any, apply to those actions, and then applies them. The DM can determine a certain action without resorting to rules: if something cannot reasonably fail, no rules are necessary and the DM simply states the outcome. If randomness is involved, calling for a certain roll is usually warranted.

So the exchange should actually go more like this:

Daxius: "I'd like to investigate Statue 1."

DM: "How?"

Daxius: "I'll look first, watching out for trap trigger mechanisms or similar. If it looks safe, I'll feel the surface."

DM: "Roll an Investigate check."

Daxius: "Solid 19."

DM: "You walk around the statue, gliding your fingers along the rough outlines of it's crisp, defined craftsmanship. You pause as your fingers make contact with the mouth. You feel warm air coming out, and instinctively look to the statue's chest. It appears to be breathing, though very faintly."

If Daxius did it as above, he made it clear he intends to touch the statue and thus gave the DM explicit permission to narrate results of touching it, for weal or woe.

If the game is always run this way, there is very little potential for meta-gaming on the second statue. It's quite natural for the exchange to follow like this:

Daxius: "Hmm... we'll get back to that in a second. Let's see what's going on with Statue 2."

DM: "Same approach?"

Daxius: Nods. Rolls... "Ugh, Investigate 6."

DM: "Hold your horses. As soon as you touch the statue, you are instantly overcome with abrasive magic, bringing your investigation to an abrupt stop. Roll a Con save."

As an alternative to "Hold your horses," the DM could have taken the Investigate roll (or asked for one, if Daxius hadn't voluteered it) as determining whether the character notices any warning signs as part of the visual inspection part of the statue. If there are such signs to be seen and the roll had been high enough to notice them, the DM would of course have stopped the action mid-progress and asked whether the player wanted the character to proceed to the touching part or not.

To summarise: Have your players state their in-world actions instead of rules usage. Always be clear on what and how is being attempted, and base your DM ruling on that. It's the DM who asks for rules to be used, not the player.

The approach above is my interpretation of how D&D-like games work best. It was largely influenced by the philosophy of Angry DM, mainly his article on Adjudicating actions (warning, strong language at end of link).

For the particular example, the description of the first statue makes it absolutely clear that investigation involves touching the statue. When the player asks to "investigate" the second they are well aware that touching is involved and the DM is not reducing their agency by assuming so.

More generally, communication is an iterative process and the onus is on the speaker to make sure the listener understands the message. It also does not occur in a vacuum, both the immediate context and the more general shared assumptions of the group are relevant. For example, it may be established either explicitly or implicitly that Investigation involves touching and Perception doesn't. For your example, the immediate context indicates that this is so: the player saying "I investigate ..." here communicates "I touch ...".

Retcons are only appropriate where the player has reasonably misunderstood the situation the DM has described. That is, the failure of communication is the DM's fault.