v0.11.0

Adds the two LLM-as-judge tools to npm: qulib_score_bug_report (grades a bug report vs a planted-bug target) and qulib_score_decisions (pivotal-decision evaluation — senior-correctness at agent decision forks). Both are prompt-injection hardened and run deterministically without a key. Documents the open-core boundary (fully open + local tool; calibrated benchmarks + hosted service are the gated tier). Publishes @qulib/core then @qulib/mcp at 0.11.0 via OIDC trusted publishing.