Adding some constant folding support for basic floating-point operations

[tannergooding]

This covers operations where it is safe to do so in the face of NaN and +0 vs -0.

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[tannergooding]

CC. @dotnet/jit-contrib, just as the title indicates this enables constant folding for floating-point where it's applicable to do. Namely this applies to NaN propagation and a few cases where operations with 0, -0, or 1 are safe for all inputs. It adds comments indicating which common scenarios we can't handle and why.

With the NaN propagation, it should be possible to also introduce (in a future PR) a minimal FloatingRange equivalent to IntegralRange and do some very basic optimizations in the space. For example, if the user has an if (float.IsNaN(x)) { return ...; } check, then there are other optimizations we can enable to reduce the total number of branches and other handling. -- I'm working on a prototype of this locally as it can have some meaningful impact to ML and other scenarios

[EgorBo]

There is a convinient API that can help with side-effects, e.g. in this case it's:

gtWrapWithSideEffects(nanCns, op, GTF_ALL_EFFECT);

it's also fine to conservatively give up on them if it doesn't affect diffs I guess

[tannergooding]

👍, didn't realize we had a helper for this. I've updated it, might get a follow up PR that looks to do the same for the integer folding.

[EgorBo]

LGTM with a few comments in gtFoldExprSpecialFloating