[front] fix: stop the new usage page from double-fetching the credit pool

[avervaet]

Description

• The new usage page's credit cards (CreditPoolCards) fetch the pool current-cycle and cycle-history endpoints themselves. UsagePage was separately fetching the same current-cycle figures again through the awu-pool-summary endpoint, only to feed the legacy pool section and the top-up dialog.
• Each of those extra requests still resolves a full session/auth context against Postgres before the handler runs, so the redundant fetch added real DB load on every pageview of the new usage page (surfaced as connection-pool acquisition delay warnings).
• Read the current-cycle figures through the same useAwuPoolCurrentCycle hook CreditPoolCards already calls, instead of the separate awu-pool-summary call. SWR dedupes identical keys, so the two call sites now share a single request instead of issuing two.
• With UsagePage no longer mounting a live useAwuPoolSummary, its post-purchase refresh-polling machinery (awuPostPurchaseRefreshState / resetAwuPostPurchaseRefreshCount and the hook's refreshInterval) had no subscriber left to poll — removed it, along with the calls to it from useAwuPurchase and useRedeemPoolTopupCoupon.
• Fixed a loading-state regression introduced by the swap: the legacy pool section's outer guard hid the whole section while awu-pool-current-cycle was loading, making its own nested spinner unreachable. It now shows during loading too.
• Applied the same fix to the poke path: PokeCreditPoolCard (poke's workspace usage tab) fetched the same heavier awu-pool-summary endpoint just to read fields already returned by awu-pool-current-cycle. Swapped it to usePokeAwuPoolCurrentCycle and removed the now-unused usePokeAwuPoolSummary hook.

Tests

• npx tsgo --noEmit and Biome pass on all changed files.
• Manual: not exercised end to end in a browser; behavior is unchanged, only the number of requests changes.

Risk

Low-to-moderate. The data-source swap itself is equivalent (same backend computation, same defaulting), but two assumptions aren't independently verified:

• The removed 5×5s post-purchase polling on awu-pool-summary assumed the balance is already consistent by the time a purchase reports "succeeded." That polling existed for a reason (likely billing eventual-consistency); removing it trades a retry safety net for a single refetch on awu-pool-current-cycle, which has no equivalent retry.
• No manual/browser verification of the purchase flow (buy credits, confirm the pool numbers and locked/unlocked state update correctly) — only type-checking and lint were run.

Recommend a quick manual check on staging (purchase credits, watch the legacy pool section and top-up dialog update; check poke's workspace usage tab) before merging.

Deploy Plan

Deploy front

[avervaet]

r? @tdraier

[github-actions]

cc-verify: violations found!

• async-network-loading-state: Legacy pool hides loading state