v2.5.0

⚠️ DO NOT USE — CRITICAL DATA-INTEGRITY BUG

This release has a silent chunk-eating bug in the streaming pipeline: when input BAM contains more than one chunk's worth of reads (default chunk_size=500), workers drop reads without producing any footprint tags. The apparent throughput (~1k reads/s) is bytes-in to an empty pipeline, not actual work.

Symptom: output BAM has 0 reads with ns/nl/as/al tags despite fast apparent throughput.

Affected versions: v2.3.0 through v2.6.2.

Fix: upgrade to v2.6.9 or later (v2.7.0+ recommended).

pip install --upgrade fiberhmm