feat(simd-split-at-semicolon): Semicolon index calculation using SIMD (~= 10%)

[manuelgdlvh]

SIMD works best when we don’t need extra assumptions or manual data alignment, and when all calculations are data simple pipeline transformations.

Thanks to the guarantees discussed in #2 (comment) about where the ; can appear (between buffer_len - 7 and buffer_len - 4), we can safely use a u8x4 SIMD search and get his max power.

With this change, the computation time goes below 700 milliseconds.

Before

After

I also made a small change in find_new_line(), related to an edge case.
When there are no more aligned-lane chunks, it is usually better to use a scalar calculation at the end instead of aligning the buffer to use SIMD.
This has no high impact, since this path is executed very few times.