These aren't mutually exclusive choices.

If someone is violently allergic to peanuts, I certainly endorse them overcoming that reaction if they're able to do so (e.g., if there's a viable cure for peanut allergies available) but I also endorse accommodating it (e.g., by not putting peanuts in their food).

Depends how reliable a signal of threat it is, and also on how we feel about sexual status markers. As others have noted, the behavior that gets labeled "creepy" covers a pretty wide spectrum -- sometimes it's a reliable indication that the person you're dealing with is sexually threatening, sometimes it's an indication of inexperience or low status but not a strong marker of threat, and occasionally it shows up due to conflicting social protocols even when both parties are high-status and nonthreatening.

The latter's straightforwardly something to overcome, or at least to recognize and route around. The former's straightforwardly adaptive. It's the "indicator of low status" category that gets ambiguous, but I don't think it's obvious that we'd be better off if our concept of sexual status was weakened or abolished.

On the other hand, if something in our culture (rather than our basic emotional machinery) is causing us to unnecessarily conflate low status with actual threat, then that also seems suboptimal; even if the subject would be rejected either way, it can't be pleasant for either party for him to be assumed dangerous. In that case the fault's in the culture, though, not in the emotional reaction.

Not if it comes from terminal values - don't wanna modify in caveats to those.

Note that it doesn't have to be a terminal value - it can be a rational response if something like P(terminal values negatively impacted | creepy) > P(terminal values negatively impacted | average) is true.