aw-sync: one-off cleanup of duplicates accumulated before #713 (synced buckets and Android staging dbs)

[ErikBjare]

Follow-up to #711 / #713. #713 stops new boundary duplicates; it does not remove what already accumulated:

• destination buckets on any desktop that pulled before fix(aw-sync): dedup boundary events on resume to prevent cursor re-imports (#711) #713 — e.g. aw-stopwatch-synced-from-poco_f8_ultra on erb-m2 holds 550 identical copies of one finished stopwatch event (plus 2 of the running:true row), …-web-chrome-… one exact duplicate;
• Android staging dbs written by a pre-fix(aw-sync): dedup boundary events on resume to prevent cursor re-imports (#711) #713 push — the phone's poco_f8_ultra/…/test.db holds 293 copies of the same event, which every fresh desktop import will still pull in once (then correctly stop growing).

Needed:

1. aw-sync dedupe [--bucket …] [--dry-run] (or part of status --fix): for -synced-from- buckets, collapse rows with identical (timestamp, duration, data) to one, reporting counts per bucket before acting. Never touches non-synced buckets.
2. For Android staging, the same on the phone's own staging db before push, or a "reset sync staging" action that drops {hostname}/{device_id}/test.db and re-stages from the local datastore (the staging is derived data). Belongs with the aw-android bump to a fix(aw-sync): dedup boundary events on resume to prevent cursor re-imports (#711) #713-carrying aw-server-rust.

The v2 format makes this class impossible (idempotency by source event id); this is the legacy-path cleanup only.

cc @TimeToBuildBob

[TimeToBuildBob]

Opened #718 for item 1 (the -synced-from- bucket cleanup): aw-sync dedupe [--bucket ...] [--dry-run], collapsing exact-duplicate events (same timestamp/duration/data) down to the lowest-id copy, over the local aw-server API so it doesn't touch sqlite files directly. Tests + clippy + fmt clean locally.

Item 2 (Android staging test.db reset/dedupe before push) is aw-android-side work and out of scope for #718 — will need its own issue there once #718 lands and the aw-android bump happens.

[TimeToBuildBob]

Status against the two items:

1. -synced-from- bucket cleanup: shipped. aw-sync dedupe [--bucket …] [--dry-run] merged in feat(aw-sync): add dedupe subcommand for pre-#713 -synced-from- duplicates #718 (a5de81b). Final pagination cuts a page at the timestamp of its PAGE_SIZE-th event and only accepts it once an older event proves the duplicate run is complete, so runs straddling a page edge (your 425–550-copy groups) are no longer split. Suggest re-running --dry-run on erb-m2 first: the count should still be 519,313. Then delete per bucket, one --bucket at a time.
2. Android staging: split out to Sync staging db: reset/re-stage to drop pre-ActivityWatch/aw-server-rust#713 duplicate copies (item 2 of ActivityWatch/aw-server-rust#717) aw-android#296 with a concrete design. Since the staging db is derived data, the proposal is a one-shot resetStaging (own-device dir only, under the existing syncInFlight guard) rather than an in-place dedupe. The one open question for you is automatic one-shot after upgrade versus a Sync Settings button; I lean automatic. It also lists what to verify first, notably that the phone's local android-test/android-unlock buckets aren't themselves duplicated, since re-staging would copy that back.

Leaving this issue open until ActivityWatch/aw-android#296 lands, since item 2 is still outstanding.