feat(amr): add --species-like wildcard filter

[thanhleviet]

Closes #33.

atb amr gains --species-like, matching the flag of the same name on atb query, so the command from the issue now works as written:

atb amr --hq-only --species-like "Campylobacter_D jej%" --format tsv --output raw_AMRFP.tsv

Pattern rule

% matches any sequence of characters. Every other character, including _, matches itself. GTDB splits genera into lettered clades (Campylobacter_D, Enterococcus_B), so an underscore has to be literal for those names to be typeable.

Partition narrowing

The AMR store is partitioned per genus, so the species filter does double duty: a row predicate and a file selector. The literal text before the first % decides which partition files can hold a match:

Pattern	Files read
Campylobacter_D jej% (literal contains a space, so the genus is exact)	Campylobacter_D.parquet, else _other.parquet
Streptococcus% (genus prefix)	every Streptococcus*.parquet plus _other.parquet
%coli (leading wildcard)	full 1.18 GB scan, with a stderr note explaining why

_other.parquet is always included for prefix patterns because it sweeps up every genus below the 10k-row partition threshold. On the real dataset that is 2,617 of 2,658 genera but only ~2% of samples (64,492 of 3,227,665), so including it is what makes the narrowing complete rather than silently lossy, and it stays cheap.

--species-like also counts as a filter for the "scan everything?" confirmation prompt, so it no longer demands --yes.

Beyond the issue

Two divergences from the documented rule turned up while wiring this in, both now fixed by routing every path through one shared internal/match package:

1. internal/query/filter.go fell back to an exact string comparison for patterns with an interior wildcard. docs/guides/query.md has been teaching --species-like "Enterococcus%faecium" as the way to search across GTDB naming variants, and that only worked when the query happened to be answered from a SQLite index rather than from parquet.

2. internal/index/query.go (atb query's index path) passed the pattern to SQL LIKE unescaped, so _ acted as a single-character wildcard. "Escherichia_coli" matched Escherichia coli. This one is a behavior change to an existing command and was outside the change set originally scoped for this issue; it is here because the docs added by this PR assert "_ is literal" for both commands, and that has to be true of both. Real-data impact is near-nil: a _ wildcard also matches a literal _, and no GTDB name has the shape Campylobacter?D, so the only thing lost is accidental matches such as "Escherichia_coli" → Escherichia coli. Called out under Fixed in the changelog.

Verification

• go build ./... && go vet ./... && go test ./... — 20/20 packages ok
• mkdocs build --strict — clean; make docs regenerated the CLI reference
• Unit tests cover partition selection, both execution paths returning identical rows for 7 patterns, and a test that corrupts Escherichia.parquet and deletes its sidecar index so reading an unnecessary partition fails loudly
• Real data, the issue's exact command against ~58.5M AMR rows: 277,638 rows in 3m09s, 922 MB RSS. Species returned were Campylobacter_D jejuni, jejuni_A, and jejuni_D (the two sub-clades an exact --species match silently drops), genus came back as Campylobacter_D confirming the partition filename derived from the pattern, and no leading-wildcard note was printed. That run had no amr/ partition directory, so it exercised the monolithic worst case; with partitions built the same pattern reads one file
• atb query --species-like "Campylobacter_D jej%" against the live 3.2M-row index still resolves in 0.46s after the escaping change