fix(bitkmer): correct reverse complement window size and length tag in minimizer

[IgnatiusPang]

Which issue does this PR close?

Fixes an algorithmic defect where bitkmer::minimizer miscalculated reverse complement values of candidate minimizers and returned an incorrect length tag, causing algorithmic divergence from sequence::minimizer.

---

Rationale for this change

1. Reverse Complement Window Size Mismatch

In bitkmer::minimizer:

for _ in 0..=(kmer.1 - minmer_size) {
    let cur = bitmask & new_kmer;
    if cur < lowest {
        lowest = cur;
    }
    let cur_rev = reverse_complement((bitmask & new_kmer, kmer.1));
    if cur_rev.0 < lowest {
        lowest = cur_rev.0;
    }
    new_kmer >>= 2;
}

bitmask & new_kmer extracts a window of length minmer_size. However, the tuple passed to reverse_complement was (bitmask & new_kmer, kmer.1), passing the parent k-mer length kmer.1 (e.g., 31) instead of minmer_size (e.g., 15).

Because reverse_complement shifts right by 2 * (32 - len), passing kmer.1 causes the reversed bits to be shifted by 2 * (32 - kmer.1) instead of 2 * (32 - minmer_size). This misaligns the reverse complement by 2 * (kmer.1 - minmer_size) bits and leaves bit-inverted junk in the lowest bits:

• Concrete Example: For sequence "AGT" ((0b00_1011, 3)) with minmer_size = 2:
  • 2-mer "GT" (0b1011 = 11) has reverse complement "AC" (0b0001 = 1).
  • Passing k=3 shifted by 58 instead of 60 bits, computing cur_rev = 0b111 (7) instead of 1.
  • As a result, bitkmer::minimizer failed to find the true minimizer "AC" (1) and erroneously selected "AG" (2).
  • This caused bitkmer::minimizer to disagree with sequence::minimizer on the exact same input (sequence::minimizer(b"AGT", 2) correctly returned "AC").

2. Return Value Length Tag

minimizer returned (lowest, kmer.1). A minimizer of length minmer_size must carry minmer_size as its length tag. Returning kmer.1 caused downstream functions like bitmer_to_bytes to read kmer.1 bases from the top, decoding prepended phantom 'A' bases (e.g. decoding a 2-mer within a 3-mer as "AAC" instead of "AC").

---

What changes are included in this PR?

1. src/bitkmer.rs:
   • Passed minmer_size instead of kmer.1 into reverse_complement.
   • Returned (lowest, minmer_size) as the function result and on early-exit bounds.
   • Guarded minmer_size == 0 || minmer_size > kmer.1 and minmer_size >= 32 exponentiation.
2. Tests:
   • Corrected test_minimizer assertion for "AGT" to expect canonical minimizer (0b0001, 2) ("AC").
   • Added test_minimizer_matches_sequence_minimizer, verifying exhaustive parity between bitkmer::minimizer and sequence::minimizer across all window lengths.

---

Are these changes tested?

Yes:

• cargo test --lib: 42 passed (0 failed).
• All integration and doc tests passed cleanly.

[audy]

this should probably just be an error. @IgnatiusPang can you change this to an assert + add a test

[IgnatiusPang]

Thanks @audy! Updated to an assert!(minmer_size > 0 && minmer_size <= kmer.1, ...) and added #[should_panic] unit tests for both minmer_size == 0 and minmer_size > kmer.1.

This also matches sequence::minimizer behavior where invalid window lengths panic. Pushed in commit 16bbca7!

(PR performed with assistance of AI)