fix(sync): reconcile cached state inspection onto current main

[KooshaPari]

Supersedes the implementation path of #849 without modifying that preserved PR or its source branch.

This branch is a clean, provenance-preserving -x replay of #849's seven commits onto current upstream main (06c6504). It retains the cached-state CLI, the non-mutating bare-gate inspection contract, explicit custody-recovery guidance, and the regression coverage that caught the prior hidden Git-object write.

Validation on this exact successor:

• make lint
• targeted and race-enabled cached-state CLI/branchsync tests
• clean baseline race batches for internal/branchsync, internal/pipeline/..., and internal/cli
• Prettier and diff checks

#849 remains preserved because its sole hosted e2e result hit a transient Git maintenance-lock template-copy race and cannot be rerun by this account. This successor requires a fresh hosted review/CI matrix and human approval before any merge claim.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the previously reported recovery-guidance and cached-state telemetry issues are fixed on the current head.

_{Reviews (2): Last reviewed commit: "test(e2e): give expanded matrix realisti..." | Re-trigger Greptile}

[kunchenguid]

Speaking as Kun's firstmate:

First look at HEAD b41e23d4 (cherry-pick successor of #849 onto current main). First-time fork CI approved (CI / docs / Guard / Require).

Diff review agrees with Greptile P1s before merge:

1. blocked_recover_explicit_verification_required sets recover_custody / axi sync --recover, but humanSyncSummary (and TUI recoverable detection) still only special-case blocked_pipeline_owned_recoverable, so cached status/TUI can hide the recover action behind generic pipeline-owned text.
2. statusFingerprint now embeds the full cached summary (branch, short HEAD, cleanliness). That expands remote telemetry beyond prior idle/run ids — please keep the fingerprint locally useful without shipping raw local branch/worktree detail in telemetry (VISION evidence custody).

Otherwise the read-only cached inspect + deferred merge-tree proof still looks like the right fix for #856. Waiting on those presenter/telemetry fixes (and green hosted CI) before merge. Please close #849 when this is the active path.

VISION.md:

• R1: aligns.
• R2: aligns (never-lose-work recovery honesty).
• R3: aligns.
• R4: aligns.
• R5: does not align until recover guidance is visible on cached surfaces and telemetry stops embedding raw local state.
• R6: aligns intent.
• R7: aligns (axi sync misses pipeline head protected by run recovery ref #856).

[KooshaPari]

Current head dcb1406 is locally validated: make lint, go test -race ./..., go build ./..., and default make e2e all pass; the e2e wrapper timeout was raised from 480s to 10m after two full measured runs (477s and 481s) proved the prior global deadline was too short. Greptile is green and all review threads are resolved. The upstream GitHub workflow matrix has not been scheduled for this fork PR head; a maintainer must approve/enable the fork workflows. After fresh hosted build, docs/guard, platform, and e2e checks pass, this remains a squash-only human-approval merge; no bypass or auto-merge is requested.

[KooshaPari]

Integration scope has been corrected: this work is now contained entirely in the KooshaPari fork. The preserved successor review is KooshaPari#5 at the same exact head dcb140615d722eef693851edf04127672e38677a. Closing this upstream proposal without merging; no branch or commit is deleted.