feat: incremental matcher rework

[saghen]

The Matcher API should be the main entrypoint for consumers. We want to do as much work up front as possible

• Copy strings to predefined 16 byte aligned widths ahead of time
  • Loop unrolling + faster SIMD loading
• Perform interleave every iteration
  • Allows us to change the number of items we operate on but requires converting score matrices to an array, TBD how slow this will be
  • If we didn't do this, how would we handle filtering? Perhaps we could do scatter/gather to combine SIMDs to only the unfiltered ones. We would then get the same performance as the one_shot matcher, or better because of the string copy

Currently the copying, prefiltering, smith waterman, and storing scores have been implemented. We don't yet calculate typos, support incremental matching or push matches to the output. Preliminary benchmarks show 67ms vs 39ms on the chromium bench. By the end of this PR, I'm hoping to be ~20% away from the one shot bench.

Shout out ii14 for all the ideas :)