Reject LUT loads that exceed the hardware address space and assemble
each 32-bit LUT word via memcpy so unaligned and partial inputs do not
read past the caller buffer. Add regression tests for oversized and
partial unaligned LUT loads.
Clear the returned handle on loopback creation failure and route
initialized objects through bitscrambler_free so channel ownership and
extra cleanup state cannot leak after partial setup errors. Add a
regression test for the failed create path.
This adds an assembler for the BitScrambler assembly language,
plus unit tests for it. It also adds the loopback driver,
which can do BitScrambler operations on memory-to-memory
transfers. Documentation is also included.