[libc] Move `__cxa_thread_atexit_impl` into proper entrypoint

As discovered in https://github.com/llvm/llvm-project/pull/224436, the function `__cxa_thread_atexit_impl` is currently defined in https://github.com/llvm/llvm-project/blob/8fe013c728a2d29722925e9bebbd454a5e3ad365/libc/src/__support/threads/thread.cpp#L146. This causes issues with this libcxxabi test: https://github.com/michaelrj-google/llvm-project/blob/4dff0fc5c826f2c6b5cb7c35aecbcc670ed07fac/libcxxabi/test/cxa_thread_atexit_test.pass.cpp#L21

Goals: Move `__cxa_thread_atexit_impl` into its own entrypoint, add tests as relevant.

Start by reading libc/src/__support/threads/thread.cpp around the current __cxa_thread_atexit_impl definition, then inspect libcxxabi/test/cxa_thread_atexit_test.pass.cpp and the linked change for the entrypoint requirements. Done means the function is provided from its own entrypoint and relevant tests are added or updated.

Written by the indexing model from the issue text.