perf: speedup parsimony ancestral reconstruction

[ivan-aksamentov]

• Simplify speedup and consensus decision in backward pass of the ancestral Fitch method: removes izip and ManyZip pythonisms, flattens and simplifies the loop
• Introduce BitSet128 data structure: a fixed-size 128-bit integer set with fast constant time lookups. It can store up to 128 chars, which should be enough for our char-based alphabets. For larger alphabet we might have a separate implementation.
• Replace BTreeSet-based StateSet implementation with BitSet128
• Replace sets in Alphabet with BitSet128 (which is now the same as StateSet)

These optimizations produce 2x speedup when running ancestral parsimony reconstruction on mpox-500 dataset (and a little more when combined with #290 and #291):

Before:

$ cargo -q build --release --target-dir=/workdir/.build/docker --bin=treetime
$ hyperfine --warmup 1 --show-output '/workdir/.build/docker/release/treetime ancestral --method-anc=marginal --dense=false --tree=data/mpox/clade-ii/500/tree.nwk --outdir=tmp/smoke-tests/ancestral/marginal/mpox/clade-ii/500 data/mpox/clade-ii/500/aln.fasta.xz'
Benchmark 1: /workdir/.build/docker/release/treetime ancestral --method-anc=marginal --dense=false --tree=data/mpox/clade-ii/500/tree.nwk --outdir=tmp/smoke-tests/ancestral/marginal/mpox/clade-ii/500 data/mpox/clade-ii/500/aln.fasta.xz
  Time (mean ± σ):      4.012 s ±  0.151 s    [User: 13.367 s, System: 0.548 s]
  Range (min … max):    3.915 s …  4.379 s    10 runs

After:

$ cargo -q build --release --target-dir=/workdir/.build/docker --bin=treetime
$ hyperfine --warmup 1 --show-output '/workdir/.build/docker/release/treetime ancestral --method-anc=parsimony --dense=false --tree=data/mpox/clade-ii/500/tree.nwk --outdir=tmp/smoke-tests/ancestral/marginal/mpox/clade-ii/500 data/mpox/clade-ii/500/aln.fasta.xz'
Benchmark 1: /workdir/.build/docker/release/treetime ancestral --method-anc=parsimony --dense=false --tree=data/mpox/clade-ii/500/tree.nwk --outdir=tmp/smoke-tests/ancestral/marginal/mpox/clade-ii/500 data/mpox/clade-ii/500/aln.fasta.xz
  Time (mean ± σ):      1.852 s ±  0.095 s    [User: 4.591 s, System: 0.523 s]
  Range (min … max):    1.768 s …  2.055 s    10 runs

One test is failing. We need to figure out ambiguous character reconstruction once and for all. But it will happen in next PR.