Support f32/f64 in native function calls

[FranciscoTGouveia]

Fixes #4642.

I would appreciate feedback on the following:

• Initially, I decided to add two test cases -- one for f32 and another for f64 -- but I later concluded that this might be redundant. If not, I am happy to reintroduce the second one;
• I have thoroughly looked into the documentation of rustc_apfloat, and the only suitable solution I have found was from_bits. I may have overlooked something, but going through the codebase, I did not find any other way to do this;
• As all operations done (to_bits() -> cast -> from_bits()) are bit-preserving, I do not think we need to worry about signaling NaNs here. Am I correct thinking this should not be problematic?

Thank you in advance for your time and input!

[rustbot]

Thank you for contributing to Miri!
Please remember to not force-push to the PR branch except when you need to rebase due to a conflict or when the reviewer asks you for it.

[RalfJung]

This looks great, thanks! Please squash the commits using ./miri squash. Then write @rustbot ready after you force-pushed the squashed PR.

@rustbot author

[rustbot]

Reminder, once the PR becomes ready for a review, use @rustbot ready.

[FranciscoTGouveia]

I believe this was a spurious failure in the CI.

@rustbot ready