First sync exhausts retries when active watcher is running

[rupertsworld]

Bug

On first sync of a large repo with an active stash watch on another machine, sync fails with "Remote main moved during push" and exhausts all 5 retries.

Cause

First sync has no local snapshot, so fetch() downloads all blob metadata and file content via GraphQL + raw API calls. This makes the window between fetch() and push() very wide. An active watcher consistently lands a commit during this window. Each retry repeats the full expensive fetch, so the window stays wide across all attempts.

Subsequent syncs are unaffected — they diff against the local snapshot and only fetch the delta, keeping the window small. If a normal sync does hit this, the next attempt self-heals.

Proposed fix

Split first sync (empty local snapshot) into two passes:

Pass 1 — pull only:

1. fetch() — download remote state (expensive, but no push follows)
2. reconcile() — compute mutations
3. Skip push — don't push snapshot.json yet
4. apply() — write remote files to local disk
5. saveSnapshot() — write local snapshot only

Pass 2 — push snapshot (immediate, automatic):

1. fetch() — fast, local snapshot now matches remote, most files skip
2. reconcile() — only snapshot.json needs pushing
3. push() — tiny payload, tiny contention window
4. saveSnapshot() — update local snapshot

Only triggers when local snapshot is empty at sync start. All other syncs unchanged.