Use rounding instructions on aarch64

[TDecking]

This extends the use of dedicated rounding instructions to the remaining rounding related functions such as trunc.
This also fixes an incorrect use of the frintn instruction, which was used for the implementation of rint instead of roundeven.

ci: allow-many-extensive

[tgross35]

I had a similar change as part of the original diff for rust-lang/libm#459. The result was that we can omit these since we will have them in core soon (rust-lang/rust#138087) and LLVM will do a reasonable lowering from there.

Regarding rint: with Rust/LLVM we don't interact with fpenv and also don't make any guarantees about it to C code that calls Rust functions, so frintn is the most simple instruction to meet our uses of rint. There is a bit of discussion at rust-lang/libm#459 (comment).