Release v2.11.0 - Checkout Lock, Per-Phase Model/Effort Config · sequant-io/sequant

What's Changed

Per-phase model/effort configuration for the claude-code driver — --models/--efforts flags on sequant run/sequant ready (#914)
Evidence-based effort escalation on quality-loop retries — --escalate-effort flag (#915)
Checkout-scoped lock for the shared working tree — sequant locks checkout <acquire|release|check|clear>, enforced by the pre-tool.sh hook (#901)
New sequant worktree resolve/verify command group (#899)

/spec's dead "Agent Spawn Rules" table replaced with the strategy actually used in practice (#927)
sequant run resolves the spec→run phase recommendation via a durable comment marker instead of ephemeral chat text (#921)
A phase-restricted run (e.g. --phases spec) no longer reports failed for lack of a PR (#920)
resolveLocksDir resolves against the shared git checkout root instead of process.cwd() — fixes a silent no-op when acquiring a lock from a subdirectory or worktree (#909)
Stealing a stale lock is now a compare-and-swap, so two sessions can no longer both win it (#908)
new-feature.sh no longer mutates the main checkout — closes the last bypass of the checkout guard (#910)
/merger and /release now participate in the checkout lock (#911)
Checkout release now requires proving you hold it, plus related lock hardening fixes (#906)
/fullsolve now creates its worktree and /exec verifies it — runs no longer execute in the main checkout or another repo's worktree (#899)

Full Changelog: v2.10.0...v2.11.0