Directory sync restore for cloud sessions handles refused links and remembers the source branch

What

--cloud starts or continues a cloud session, where Claude Code runs on Anthropic's machines instead of yours. A background worker keeps your project folder in step with that session (directory sync). When a cloud session restarts, the worker restores the earlier environment. This release changes that restore in several ways:

• When the worker fetches its own links and gets an "unauthorized" or "unavailable" answer, it now reports a separate lane_refused result. Those links are marked unusable, and a restore stopped this way reports "unreadable". Before, an unauthorized fetch was treated as still waiting.
• Whether the storage service is down (store_down) is now decided at the fetch itself.
• A git_error while receiving is attempted once instead of being retried in a loop.
• While waiting on its own journal, the worker now writes a log line every minute.
• The worker now works out which branch on your laptop the restored starting point came from. The source can be the laptop's note, the earlier note's basisBranch, both agreeing, or unknown, and it saves basisBranch with the restore.
• If the earlier session's links had stopped (linksStopped), the restore is now partial and marked not_stored.
• Restore reports gain branch_at_basis_from and earlier_links_stopped.

All of this is tied to the gate tengu_dir_sync_git_worker_restore, a remotely controlled switch. Nothing has been read about this gate, so which way it is set is unknown.

Why

These changes affect how a restarted cloud session gets back your earlier work. A refused link now fails quickly instead of retrying or waiting, and the restored work is more often linked to the right branch on your laptop.