turbo-picard 0.1.13

Native execution and sorting

• Use a heap for BAM external-merge selection instead of scanning every active
  run. Preserve the existing comparator and cross-run tie order.
• Track owned spill files through intermediate merges and failures; reject
  truncated spill record headers and account for sorting metadata in memory
  budgets.
• Reduce optical-coordinate parser allocations and read-name cloning; protect
  coordinate-distance and neighbouring-grid arithmetic against overflow.
• Add TURBO_PICARD_REQUIRE_NATIVE to prohibit both explicit and automatically
  discovered Java fallback during native-only evaluation.

Workflow and agent interfaces

• Add command-scoped capabilities --json --command <Command> discovery.
• Provide executable argument arrays and command-specific output roles in trial
  JSON, with explicit syntax, option-support and input-inspection boundaries.
• Stream or externally sort alignment comparisons without sampling or dropping
  records. Preserve failed evaluation work and refuse to replace existing
  evidence; --discard-work cleans only a successful current run.
• Reject NaN-versus-number metric mismatches and handle missing optional tags
  deterministically in duplicate-semantic comparisons.

Documentation and evidence

• Clarify the README, landing page, agent guidance and real-data evaluation
  instructions; improve keyboard, mobile and reduced-motion behaviour.
• Retain absolute timing, repeat-count and workload metadata in benchmark JSON;
  correct the even-count suite median from 99.56x to 99.51x using unchanged
  August 14 saved logs. This release does not claim a new production-scale
  benchmark campaign or universal superiority over other tools.
• Include the reproducible validation-helper memory benchmark (one million
  synthetic SAM records); its results do not measure native MarkDuplicates.
• Fetch tags for public adoption audits, discover new Python regressions in CI,
  and fix strict Sphinx heading validation.

Bioconda acceptance and version-specific archival DOI assignment are separate
from the GitHub/PyPI/container release. Cite the exact software version and
input-specific evidence used.

Install: python3 -m pip install --upgrade turbo-picard==0.1.13

Container: ghcr.io/dnncha/turbo-picard:0.1.13