chore(release): 0.7.6 — digest-pinned, honesty-classified confined baseline

[hartsock]

Summary

Release prep for 0.7.6: bumps the workspace version 0.7.5 → 0.7.6 (package + the 17 internal =0.7.x pins + Cargo.lock), cuts CHANGELOG [0.7.6], and banks the confined (OCAP-on) tb-30 scoreboard under the honesty-classified, digest-pinned methodology.

Confined champions: ornith 36.7%, qwen3.6 26.7%, qwen3-coder 13.3%, kimi-linear 10.0%, glm-4.7-flash 3.3%. Three models were quarantined for under-engagement (granite4.1, nemotron-3-nano canonical + bare) — tracked in #1500.

This is a re-baseline under stricter measurement, not a raw champion-defeat; tb-30 is a 30-task sample (~±10pp variance). Parity (OCAP-on ≈ off) is pursued forward via pre-granted permissions, not gated here.

Test plan

• python3 scripts/eval/bench_scoreboard.py --self-test → OK
• cargo update --workspace resolves cleanly (lock synced to 0.7.6)
• CI: full just check on this PR

Out of scope

• Observability-contract model_digest emission from newt solve (folds into the feat(bench): newt bench — matrix orchestration, one model at a time (#1490) #1491 re-scope).
• Re-adjudicating the 3 quarantined models (Benchmark validity: disambiguate chat-template / tool-call-parse artifacts from genuine model weakness #1500).

Refs #1500.

---

Note

Low Risk
Version and documentation/eval-data only; no application logic, auth, or runtime behavior changes.

Overview
Release 0.7.6 bumps the workspace and all lock-step internal crate pins from 0.7.5 to 0.7.6 (Cargo.toml, Cargo.lock) and adds a CHANGELOG [0.7.6] entry documenting the confined (OCAP-on) tb-30 re-baseline under stricter measurement (≥25/30 tool-call engagement, GGUF model_digest pinning, quarantine for under-engaged models).

Benchmark artifacts extend scripts/eval/bench-results.jsonl with model_digest on key OCAP-on runs and new confined results for kimi-linear_48b and glm-4.7-flash. The auto-generated README scoreboard and bench_scoreboard.py header now describe 0.7.6 as establishing that honesty-classified baseline and note that OCAP parity is pursued forward, not gated on this release.

^{Reviewed by Cursor Bugbot for commit 2ffb69b. Configure here.}