performance: add attributes [[clang::coro_await_elidable]] / [[clang::coro_await_elidable_argument]]

[tzcnt]

[[clang::coro_await_elidable]] [[clang::coro_await_elidable_argument]]

These attributes should allow the compiler to perform HALO, hopefully leading to a good performance win.

Note that these attributes are only available in Clang 20, so detection for older versions / other compilers will need to be added.