Fix saturated_cast double to int64_t conversion near overflow

When converting double values near int64_t::max() (e.g. 2^63 - 8),
std::numeric_limits<int64_t>::max() cast to double rounds up to 2^63.
This causes the upper bound check to falsely succeed, leading to
undefined behavior or negative results upon conversion.

This is fixed by using PowerOfTwo to perform exact boundary checks
for floating-point sources.

Fixed: webrtc:42220971
Change-Id: I70e455566a6e3d6999f77ab31f3675435f542714
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/491740
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Commit-Queue: Tomas Gunnarsson <tommi@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#48234}