This is the first step in preparation for:
https://discourse.llvm.org/t/rfc-make-clang-builtin-math-functions-constexpr-with-llvm-libc-to-support-c-23-constexpr-math-functions/86450

Fix #98393

Context: https://github.com/llvm/llvm-project/pull/87017

- Add proxy header `libc/hdr/math_macros.h` that will:
  - include `<math.h>` in overlay mode,
- include `"include/llvm-libc-macros/math-macros.h"` in full build mode.
- Its corresponding CMake target `libc.hdr.math_macros` will only depend
on `libc.include.math` and `libc.include.llvm-libc-macros.math_macros`
in full build mode.
- Replace all `#include "include/llvm-libc-macros/math-macros.h"` with
`#include "hdr/math_macros.h"`.
- Add dependency to `libc.hdr.math_macros` CMake target when using
`add_fp_unittest`.
- Update the remaining dependency.
- Update bazel overlay: add `libc:hdr_math_macros` target, and replacing
all dependency on `libc:llvm_libc_macros_math_macros` with
`libc:hdr_math_macros`.

We compute atan2f(y, x) in 2 stages:
- Fast step: perform computations in double precision , with relative
errors < 2^-50
- Accurate step: if the result from the Fast step fails Ziv's rounding
test, then we perform computations in double-double precision, with
relative errors < 2^-100.

On Ryzen 5900X, worst-case latency is ~ 200 clocks, compared to average
latency ~ 60 clocks, and average reciprocal throughput ~ 20 clocks.