This change replaces inline x86 intrinsics with generic versions that compile
for both x86 and ARM depending on the target arch.
This change does not enable the accelerated crc memcpy engine on ARM. That will
be done in a subsequent change after the optimal number of vector and integer
regions for different CPUs is determined.
PiperOrigin-RevId: 562785420
Change-Id: I8ba4aa8de17587cedd92532f03767059a481f159