Support for converting to and from `f16` and `f128`

Would it be possible to add conversions between f16/f128 and float types? This would be great for testing f128 math operations that are slowly being added. Everything would need to be gated behind a nightly or similar Cargo feature.

I don't know what the bare minimum functionality is here but most things should at least compile now on the latest nightly. Notable exceptions for Tier 1 platforms:

• Symbols for {f16,f128} <-> int as casts are not yet available on all platforms, so there may be linker errors if these are used on anything but Linux.
• f128 division is not yet available, linker errors like the above
• 32-bit x86 has ABI issues for f128 https://github.com/llvm/llvm-project/issues/77401
• MinGW has ABI issues for both f16 and f128 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115054

It looks like CI might be largely MinGW so the ABI issue on that platform is unfortunate. For now it is probably okay to only cargo check without any runtime tests.