Persist the per-folder root ETag across restarts (#195)

[gnacho]

Closes #195.

Problem

The root ETag gate (#189, v0.2.10) skips the full nextcloudcmd reconciliation when the remote tree is unchanged, but the per-folder ETag lived only in the engine's in-memory etag_slot. After an app restart the slot was empty, so the first RemoteInterval of every folder re-scanned fully even with zero remote changes. On the large folders (Fotos ~65k files) that is a full tree re-scan on every launch.

Change

Persist the ETag per folder and seed the engine from it:

• New src/core/etag_store.rs: read_etag(folder_id) / write_etag(folder_id, etag) stored at state_dir()/etags/<folder_id>. Kept outside the synced folder - writing inside it would be seen by the FsWatcher and could requeue the folder (the Startup pass and queue discipline: folders re-run immediately instead of one clean pass each #181 vector). Reads/writes are best-effort; a missing/stale value degrades to "changed", which reconciles (the safe direction).
• SyncEngine::new seeds etag_slot from the persisted value.
• On a periodic-interval run that records a new ETag, the engine persists it (etag_store::write_etag) and fires an optional on_etag_change callback.
• RuntimeConfig gained remote_etags: HashMap<String,String> (serde default + migration parse) for any future wiring, though the engine persists via etag_store directly.

Testing

4 new tests (etag_store, hermetic with injectable base dir) + existing engine ETag tests. Gate: 705 tests, clippy clean, fmt applied.