feat: sync imagine quota from upstream after generation

[scvxzf1]

Summary

Imagine image/video quotas (image_pro / image_edit / video / video_720p) were read-only snapshots that never refreshed after generation, so Accounts → Grok Web showed stale quotas even though real consumption was happening. The per-generation refresh path was blocked end-to-end:

• imagine models had an empty catalog Mode, so QuotaMode() returned "";
• QueueQuotaRefresh rejected imagine modes (whitelist of console/weekly/chat only);
• runQuotaRefresh force-switched every non-console mode back to weekly;
• SyncQuotaMode only supported weekly and /rest/rate-limits, not /rest/media/imagine/quota_info;
• the image/video gateways fed effectiveQuotaMode (always weekly for web) into QueueQuotaRefresh, so imagine modes never reached the refresh queue.

This PR wires imagine models into the per-generation refresh pipeline so a successful imagine generation triggers an upstream /rest/media/imagine/quota_info sync for that account.

Changes

• catalog.go — assign image_pro / image_edit / video Mode to imagine models so QuotaMode() returns the imagine mode. (grok-imagine-image-lite keeps fast — it uses the chat endpoint.)
• quota.go — SyncQuotaMode gains an imagine branch that calls syncImagineQuota and returns the requested mode's window; full SyncQuota now retains imagine windows alongside the weekly snapshot instead of dropping them.
• service.go — new isWebImagineQuotaMode (reuses imagineQuotaBlockModels keys); QueueQuotaRefresh accepts imagine modes; runQuotaRefresh skips the force-weekly override for imagine modes; refreshQuotaMode applies applyImagineQuotaCooldown after persisting imagine windows.
• image.go / video.go — pass the catalog imagine mode (image_pro / video) to QueueQuotaRefresh instead of the always-weekly effectiveQuotaMode / lease.QuotaMode.
• pricing.go — extend the official image-cost estimator to imagine-x-1.
• model_repository.go — add imagine-x-1 route defaults.
• account_repository.go / repository/account.go — OverwriteModelQuotaBlock to align cooldown to the upstream authoritative nextAvailableAt.
• account-quota.tsx / i18n — surface imagine quota windows in the Accounts UI as a dedicated row (remaining/total, exhaustion + cooldown state, reset tooltip) with localized labels.

Routing is unchanged: getRoutingQuotaWindows keeps weekly first in its ORDER BY and the per-account map keeps the first row, so candidate.QuotaWindow and effectiveQuotaMode stay weekly and selection is unaffected. The catalog Mode only feeds QuotaMode() for the finalize refresh call.

Verification

Deployed and tested end-to-end against a live Grok Web pool (imagine-x-1):

• Generation 1 (account 11249): image_pro synced_at advanced 18:21:49 → 18:55:12 (≈1–2s after generation).
• Generation 2 (account 11251): image_pro 8/8 → 6/6, synced_at 18:21:49 → 19:00:09, reflecting the two upstream consumption events.
• Container logs confirmed POST /rest/media/imagine/quota_info fired after each generation and quota_refresh_events reported outcome=success.
• go build ./... and go vet clean; affected package tests pass (internal/infra/provider/web, internal/application/account, internal/application/gateway); frontend pnpm typecheck clean.

Test plan

• Run go test ./internal/infra/provider/web/... ./internal/application/account/... ./internal/application/gateway/...
• Generate an image via imagine-x-1 and confirm the used account's image_pro window synced_at advances and remaining reflects the upstream value in Accounts → Grok Web.
• Generate via grok-imagine-image-lite (fast mode) and confirm the existing fast-quota decrement path is unaffected.
• Exhaust an account's image_pro (remaining 0) and confirm a ModelQuotaBlock cooldown is written for imagine-x-1 / grok-imagine-image-quality.
• Confirm weekly quota and account selection behave unchanged before/after.