feat: add Amazon Quick local token usage provider

[solangegamboa]

Summary

Add local token-usage support for Amazon Quick (Quick Suite — Amazon's desktop AI assistant for macOS/Windows) through the existing provider-agnostic aggregation pipeline.

• Discover ~/.quickwork/sessions/sessions.db (single-profile installs) and ~/.quickwork/profiles/*/sessions/sessions.db (enterprise/team installs), plus optional custom scan roots, de-duplicating resolved database paths.
• Read turn-level usage with read-only SQLite connections. Only usage metadata is selected (event id, timestamp, token_usage, model); conversation bodies (full_content, session_messages.content) and credentials are never read.
• Parse real per-turn counts from message_assistant_complete events, whose data JSON carries a token_usage object with input_tokens / output_tokens / cache_read_tokens / cache_write_tokens. This is source-recorded usage, not the byte estimate Kiro falls back to.
• A single database holds every session (the session_id column separates them), so multi-session totals are the natural sum of all events — no per-session file scanning. Multiple profiles are all scanned.
• Sit on the shared LocalAdditionalUsageCache like Kiro/Aside: one scan per refresh feeds daily and enrichment, and each scan merges with previously-seen entries (dedupKeepMax(existing + loaded)). Deleting a session removes its events, so a plain rescan would make today's total go down; the merge keeps already-counted usage until the cache resets.
• The reader never throws. Unreadable databases (a foreign sessions.db under a custom root, a busy WAL recovery) are skipped; an all-failed rescan returns [], which the merge treats as "nothing new" rather than zero usage.
• Entry ids carry the file's inode so a recreated profile cannot hide its new events behind cached ids.
• Register QUICKWORK_HOME as a location override and include Amazon Quick in daily/weekly/monthly/active-block aggregation while preserving existing providers.
• Add synthetic SQLite tests (real session_events schema) and update the provider-registration and tab-layout guards.

Type of change

• New feature

Data semantics and limitations

• Usage is attributed to the local date of the turn's completion timestamp, so tokens land in the day / 5-hour block they were generated in.
• Zero-token turns (aborted before a response) are dropped, so they never form an empty active block or a Quick tab with nothing in it.
• The token_usage.model field is Quick's plan-mode label ("balanced", etc.), not a public price-table model, so cost is marked unavailable rather than priced against an unrelated model.
• A deleted session's already-counted usage stays counted until the scan cache is dropped (Settings save, month rollover, relaunch), matching Kiro/Aside behavior.
• This adds local usage accounting, not subscription-limit tracking.

UI changes

None. The existing provider tab and aggregates display Amazon Quick automatically once data is present.

Validation

• swift build and full swift test pass locally: 1088 tests, 12 skipped, 0 failures. The 10 new AmazonQuickUsageTests cover bucket parsing, multi-session summation in one DB, multi-profile + single-profile summation, the type filter, timestamp→local-day attribution, zero-token drop, unreadable-DB skip, malformed-JSON skip, provider registration + custom-root support, and shared-scan enrichment totals.
• Read-only probe of a live Amazon Quick store confirmed the parser reproduces the SQL reference exactly: overall 502,798,367 tokens across 9,035 usage entries (9,042 message_assistant_complete rows minus 7 zero-token aborted turns) and today's 1,580,845, matching a json_extract sum over the same events. No personal database or diagnostic snapshot is included in this PR.

Checklist

• swift build and swift test pass locally
• PR title and description are written in English
• No UI source changes; the existing provider tab and aggregates display Amazon Quick
• No copyrighted assets, secrets, or private tooling references are committed
• Tests were added or updated for this change