Expand N-handling test coverage + remove dead has_r2 guard

[joshfactorial]

Summary

Follow-up to the v4.6.0 N-handling work (#313, released via #314). Closes coverage gaps and a small dead-code finding from reviewing the release PR. Mostly test-only, plus one behavior-preserving cleanup.

Tests

• finalize_read_and_write threading — the new n_handling parameter was never exercised end-to-end. Added tests that the default (exclude) leaves a literal N in the emitted FASTQ and that telomere reaches convert_masking and fills it.
• Paired path — added a positive control (both mates clean → fragment kept) and extended the drop test to cover read1-in-gap, not just read2-in-gap.
• GC-bias sampling path — the prior end-to-end exclusion test only covered the uniform sampler; the GC path has its own _filter_n_regions call sites. Added a GC-path test (non-uniform model, asserted via not gc_model.is_uniform).
• Options validation — added checks for the n_handling choice schema and n_max_fraction [0,1] range, and asserted both defaults in test_default_values.
• Stale comment — test_check_and_log_error_choice_valid claimed the choice validator is unused by any production option; n_handling now uses it. Corrected.

Cleanup (production)

• Removed the dead has_r2 guard in _filter_n_regions. In the paired branch read2 is always a real in-bounds window ((e - read_len, e) with e >= read_len + 10), so has_r2 was always True; the (0, 0) placeholder only occurs single-ended, which never reaches that branch. Behavior-preserving (covered by the paired N-filter tests); also drops a misleading "filtered upstream" comment.

Testing

pytest tests/test_read_simulator/ → 402 passed. gitnexus detect-changes → low risk.

🤖 Generated with Claude Code