Adding reverse-order SWAR search backends

StringZilla currently implements several SWAR search optimizations for needles of different length: `_sz_find_2byte_serial`, `_sz_find_3byte_serial`, `_sz_find_4byte_serial`. Those currently lack reverse-order variants and should be implemented to guarantee the same level of throughput on non-SIMD devices in reverse-order operations.

Implementing those is fairly strait-forward, and can be a great first issue for people used to C programming.

Locate the existing _sz_find_2byte_serial, _sz_find_3byte_serial, and _sz_find_4byte_serial SWAR search implementations and their reverse-order callers. Use the forward variants to understand the expected behavior, then verify that reverse-order operations have corresponding implementations for all three needle lengths and preserve non-SIMD search throughput.

Written by the indexing model from the issue text.