Transcriptome preset (-x tx), 2.3-2.8x faster than minimap2 for quantification - #27

…hors dropped before chaining

- -x tx: reads to a transcriptome for quantifiers (oarfish, NanoCount): minimap2 map-ont
  seeding (k15 w10, no occurrence range, -g5000) and up to 100 secondary alignments, with the
  dna preset's chaining and DP otherwise.
- The ALT-contig count was recomputed for every read by scanning all targets (twice per read):
  32 of 35 post-chain seconds on a 534k-transcript index. It is now counted once per index.
- Anchors with no other anchor of the same reference/strand within the widest chaining distance
  are dropped before chaining (they can only form a one-anchor chain); the rescue's density test
  keeps counting all anchors. Output identical (DNA genome-wide sample 9bc7ff3e..., all golden
  hashes); DNA CPU -1.2%, transcriptome chaining -8%.
- Region building copies chain anchors only for regions that survive the secondary filter.
- Golden test tx_sam.

HG002 200k reads vs minimap2 -ax map-ont -N 100 (GENCODE v49 transcripts): cDNA 185 -> 81 s,
direct RNA 106 -> 37 s; same reads mapped; simulated reads at or above minimap2's accuracy;
oarfish gene counts r = 0.9995 / 0.99997.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>