Optimize BitfieldUnit get/set with byte-wise operations

Replace bit-at-a-time loops in get(), set(), raw_get(), and raw_set()
with byte-wise shift/mask operations. This significantly improves
performance for multi-bit field access.

The new implementation reads/writes bytes in chunks and uses shifts
and masks to extract/insert values, rather than iterating over each
bit individually.

For big-endian targets, the algorithm reverses bits within bytes and
the final value to maintain the same semantics as the original
implementation.

Author: Jeremy Fitzhardinge