nx_align_match_stranded.nx

nx_align_match_stranded.nx -- strand-aware seed matcher. license_tier: INDEPENDENT_REDERIVE genealogy_id: international-research-sources/li-2018-minimap2-section-2.1 G1.3c of NISHI_GENOMICS_SUBSTRATE_ROADMAP.md. Real-world reads arrive from either strand of the reference; a forward-only matcher misses 50% of true alignments. This primitive matches BOTH the forward-query minimizers and the reverse-complement-query minimizers against the reference, tagging each emitted pair with strand orientation. Strand semantics: q_fwd_* -- minimizers from the query as given q_rc_* -- minimizers from the reverse-complement of the query (the caller pre-extracts these via dna_revcomp + minimizer_extract) Match emits forward-strand pairs first (tag = NX_STRAND_FWD = +1), then reverse-strand pairs (tag = NX_STRAND_REV = -1). Within each strand, emission follows minimizer_match's outer-q / inner-r order so chain DP sees a deterministic stream. Why canonical-k-mer minimizers are NOT enough on their own: Canonical k-mers are strand-invariant by construction (the smaller of forward and revcomp wins), so the SET of canonical minimizer VALUES is identical between a query and its revcomp. But the POSITIONS differ: a query base at position i corresponds to revcomp-query base at position (n - 1 - i). Strand info lives in this position-mapping, not in the value. Therefore we run TWO matches and tag the output stream. q_pos semantics in the output: Forward pairs: q_pos refers to the original query's coordinate (where the k-mer starts in the query as given). Reverse pairs: q_pos refers to the REVCOMP-QUERY coordinate. Downstream chain DP and SW-extension are responsible for un-mapping reverse-strand q_pos back to original-query coordinates when needed (see G1.4b nx_alignment_record). API: