feat(sync): reconcile prompt variation changes

[ctawiah]

Context

The manifest records the last state accepted locally and in LaunchDarkly. Comparing that baseline with both current sides lets sync distinguish a local edit, server edit, deletion, and true conflict without adding a backend sync endpoint.

What changes

• Plans deterministic create, update, archive, pull, delete, manifest, conflict, and error actions.
• Renders structured plaintext, Markdown, and JSON review output.
• Revalidates local and server fingerprints immediately before mutation.
• Applies server writes through existing variation APIs and local writes through the workspace store.
• Advances the manifest only for successful outcomes.

support/default
  Action: Update LaunchDarkly from the local file

Review focus

• Does every three-way state choose the expected action?
• Can an action write after either side has changed?
• Does partial success advance only resources that converged?

Verification

• go test ./internal/sync/...
• git diff --check

Related changes

Review the stack in this order:

1. Sync foundations
2. Prompt reference workspaces
3. Prompt sync manifests
4. Local prompt workspaces
5. Bootstrap prompt workspaces
6. Link and detach prompt resources
7. Prompt reconciliation
8. Prompt sync orchestration
9. Prompt sync command

---

Note

Overview
Adds internal/sync/prompt, the core library for reconciling prompt variations between local files, LaunchDarkly, and the sync manifest.

Planning compares manifest baseline fingerprints to current local and server state and picks deterministic actions (push, pull, create, archive, delete, manifest bookkeeping, conflict, or error). First-sync and upsert behavior are handled separately from tracked three-way merges.

Review output renders plans and post-sync outcomes as plaintext, Markdown, or JSON, including colored unified diffs (side-by-side on wide terminals) via go-difflib, with readable placeholders for missing or archived resources.

Execution re-checks fingerprints immediately before each mutation, applies server changes through existing variation APIs and local changes through the workspace store, verifies local writes, and on ambiguous API failures re-reads LaunchDarkly before treating a write as success. The reviewed manifest is copied so only successful resources advance the returned manifest; failures are reported per resource.

Tests cover the three-way action matrix, first-sync rules, manifest immutability, and JSON/diff messaging.

^{Reviewed by Cursor Bugbot for commit 005bed3. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Stale Bugbot comment from a previous run.

[nieblara]

I think this seems fine, but the destructive actions would be executed from that forced-auto mode for applying changes, not sure if we want that!

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 7cc8f21. Configure here.}

[ctawiah]

I think this seems fine, but the destructive actions would be executed from that forced-auto mode for applying changes, not sure if we want that!

Good point. Thought about this a little bit more and the intended behavior should be for watch mode to auto-apply non-destructive changes, but still prompt for confirmation before archive or delete actions. I created a ticket to track that follow-up.