v0.1.1

[0.1.1] - 2026-05-15

Added

• --features accepts multiple BEDs, scored in one cell-read pass. Each BED
  writes its own triplet <prefix>.<label>.{cell_feature,feature,pair_counts}.tsv.gz
  (<label> is the BED basename). A single --features keeps the old paths.

Changed

• Streamed scoring: rows written per cell, aggregates updated as it goes. Peak
  memory is bounded by threads x one cell-and-feature-set batch, not the full
  cells x feature-sets x features matrix.
• Cells scored in parallel. cell_feature/pair_counts row order is now
  cell-interleaved (key by cell_id); feature.tsv.gz stays sorted.
• Workflow: CRC/Ecker/Argelaguet feature rules run amet once per cell-group
  combo. Output names changed from <annotation>_<combo>.* to
  <combo>.<annotation>.*.
• Makefile: CORES 16 -> 40, ULIMIT_KB 200 GB -> 100 GB per process.

Performance

• Pair counting rewritten: compact observed-CpG window, all lags in one sweep.
  Cost scales with observed CpGs, not the feature's total CpG count.

Fixed

• --i-max-lag 0 is rejected by the CLI; it no longer panics (lag 1 is needed
  for JSD).
• features_label strips .bgz and .bed.bgz.
• setup_barbara_links.sh and sync_from_barbara.sh resolved the repo root one
  level too shallow, placing results/ symlinks under workflow/.