Since this version allows new things, it's a good time to get costs
aligned better.

b+ and b- now work directly on the big-endian bytes, a word at a time,
rather than converting to and from big.Int. They are 3-4x faster, so
their v14 cost is lowered to 2 + 1 per 32 bytes of the longer input.

b|, b&, b^, and b~ now work a word at a time instead of a byte at a
time, about twice as fast, which brings them under the cost target at
large sizes without changing their costs.

Property tests compare the new code with big.Int and with byte-at-a-time
versions.

Co-authored-by: mangoplane <devmon1@protonmail.com>
Co-authored-by: mangoplane <104189015+mangoplane@users.noreply.github.com>