Release BlackSTAR 1.1.0

[justinblethrow-cloud]

Purpose

Prepare and release BlackSTAR 1.1.0 by promoting the Q02 compatibility
hardening, broadening workload coverage, and completing portable dual-variant
release packaging.

Origin Classification

• Inherited upstream behavior or fix
• BlackSTAR regression fix
• BlackSTAR-specific feature
• Documentation, test, or maintenance only

Compatibility and Risk

• Retains official STAR 2.7.11b as the compatibility oracle and genome format
  2.7.4a for conventional indexes.
• Retains the wrapper-compatible STAR --version token
  2.7.11b-blackstar.3; structured metadata reports BlackSTAR 1.1.0.
• Makes TranscriptomeSAM primary selection deterministic across worker
  schedules. The complete transcript alignment set, genomic BAM records,
  counts, junctions, and later inherited random-stream position are preserved.
• Overlay and Delta remain BlackSTAR-specific and require
  --genomeLoad NoSharedMemory.
• Release artifacts are separate baseline x86-64 and AVX2 variants with
  explicit ISA and result-equivalence checks.
• No external deployment is authorized by this pull request.

Validation

• Dedicated-node exact source-candidate qualification: 29/29 local gates.
• Specialized official-STAR differential matrix: 12/12.
• Focused ASan/UBSan scripts: 11/11.
• GenomeInsert hardening: 32/32, including stock-STAR compatibility.
• Q02 public workload comparisons: 36/36 pair-level output oracles.
• Full CHM13+ERCC substantive index identity: 14/14.
• Architecture renderer: 29 canonical figures exact.
• Two clean release roots compare every baseline and AVX2 product
  byte-for-byte, including uniquely named standalone release assets.
• Protected GitHub checks required before merge: full acceptance, GCC, Clang,
  STARlong, Ubuntu 20.04/glibc 2.31 portability, and C++/Python CodeQL.

Bounded receipts:

• docs/benchmarks/Q02-cross-workload-20260726/
• docs/benchmarks/RCQ-1.1.0-source-20260726/

Performance Evidence

Q02 confirms that the cumulative BlackSTAR gains generalize to multiple common
paired, two-pass, junction-filtered, chimeric, sorted-BAM, and long-read
workloads. The exact replicated values and variability are retained in
docs/PERFORMANCE.md.

The single-end timing series failed its variability gate and does not support a
speed claim. STARsolo and TranscriptomeSAM are noninferior-only results.
STARlong retains the documented symmetric seed-limit caveat.

Documentation

• Tests would fail without the change
• User-facing behavior and changelog are updated
• Generated architecture artifacts are synchronized
• No private data, credentials, or customer information is included