Rollup merge of #159954 - valentynkit:float-conversions, r=JohnTitor

core: implement float conversion methods

Implements the accepted ACP rust-lang/libs-team#810, adding float conversion methods on `f16`, `f32`, `f64`, and `f128`:

- `cast` converts to the target float type (`self as Flt`, IEEE rounding).
- `to_int_saturating` converts to an integer with `as` semantics: saturating, and `NaN` to 0.
- `to_int_checked` truncates toward zero and returns `None` on `NaN`, infinity, or out of range.
- `to_int_strict` is `to_int_checked().unwrap()`.

Per the libs-api meetings on the ACP, `to_float` is named `cast` and `to_int_saturating` maps `NaN` to 0 rather than panicking (which the team noted makes rust-lang/libs-team#831 redundant).

The integer-to-float `cast` from the ACP is left out: it shares the `cast` method name on integer types with the integer-to-integer `cast` in rust-lang/libs-team#811 (being unified with the other integer cast methods in rust-lang/libs-team#833), so it needs a decision on how those fit together. Happy to follow up once that settles.

Tracking issue: rust-lang/rust#159913

r? libs