1.66 regression -- non-optimized compiler-builtins by default

[Mark-Simulacrum]

We missed backporting #103732 to 1.66, so for 1.66 only the bad behavior is still there.

If we do have a point release let's try and include the revert there.

cc @pietroalbini

[apiraino]

@Mark-Simulacrum I think this will probably land on the next stable without backport (see zulip). So probably I'd remove I-prioritize (or maybe close the issue, idk).

@rustbot label -I-prioritize

[Enselic]

Triage: No objections to the proposal to close, so let's close.