[llvm] [KnownFPClass] Add output/input denormal helpers (PR #225575)

[Eric Ross via llvm-commits]

ZERICO2005 wrote:

Only `atan2` uses the new helpers in this PR. This was done because `atan2` has no denormal mode bugs, has extensive denormal mode tests/coverage, and will not conflict with other PRs.

I have tested the new helpers in this draft PR (replacing `isKnownNeverLogicalZero(Mode)` with `isKnownNeverZero()` and etc to verify the correctness of the refactor):
https://github.com/llvm/llvm-project/pull/224795

Otherwise, I plan to add the new `applyInputDenormalMode`/`applyOutputDenormalMode` helpers to other functions after these PRs merge:
- https://github.com/llvm/llvm-project/pull/219114
- https://github.com/llvm/llvm-project/pull/219303
- https://github.com/llvm/llvm-project/pull/219700
- https://github.com/llvm/llvm-project/pull/219837
- https://github.com/llvm/llvm-project/pull/220148
- https://github.com/llvm/llvm-project/pull/215094
- https://github.com/llvm/llvm-project/pull/214987


https://github.com/llvm/llvm-project/pull/225575