Improve floating point documentation about platform guarantees

[golddranks]

As summarized in #72599 (comment), floating points have both some 1) platform-specific behaviour 2) surprising behaviour because of LLVM guarantees (or lack of thereof). Issue #73328 seems to be about improving the documentation about the LLVM-induced stuff, but mention about platform-specific behaviour also seems to be missing in some documentation items, so add some explanations/disclaimers about that.

[golddranks]

I'm working on this currently.

[RalfJung]

What is this issue still tracking? Some improvements landed in #95483. The issue description as it is is too vague to be actionable.

[golddranks]

Agreed! Closing this.