gh-115480: Type / constant propagation for float binary uops

[lazorchakp]

WIP

• Issue: Type/constant/value propagation for BINARY_OP #115480

[lazorchakp]

@Fidget-Spinner - opening this as a draft because I still have a few questions and it's waiting on some code from #115507.

• Currently, the main changes are essentially untested. None of the tests actually hit the if (is_const(left) && is_const(right)) case. I get the feeling that this is expected for now, but it's entirely possible that I've just written bad tests.
• how much do we care about runtime performance of the tier 2 optimizer? Should I prefer safety or speed? For example: PyFloat_AsDouble vs PyFloat_AS_DOUBLE
  • related: I was wishing for something like _PyFloat_Add, but ended up performing each calculation manually; this feels unsafe

[lazorchakp]

Also, I'll be on vacation until Tuesday and may be slow to respond. Please feel free to use / modify this code however you see fit if you're waiting on it

[Fidget-Spinner]

Nice work! This is pretty close, I will just push some changes since you're on holiday. Some points for possible consideration are that the tests should test for the float guard being eliminated as well.

[lazorchakp]

Thanks for finalizing and merging this 🎉