feat(frontend): fix and harden order recovery from coordinator after reload

[dykdee]

Summary

The frontend already attempted to recover pending/refundable swaps from the coordinator on mount (TransactionHistory.tsx), but the request was silently broken: it queried /api/orders/history?eth=...&stellar=..., while the coordinator's route (coordinator/src/server/routes/orders.ts) only accepts a single address query param (matched against either side of the order via src_address OR dst_address). Every recovery request hit the address_required 400 path, was swallowed by the catch block, and silently fell back to local-only state — so recovery never actually worked after a reload in production.

This PR fixes that bug and hardens the recovery path per the issue's acceptance criteria.

Changes

• frontend/src/lib/orderRecovery.ts (new) — small, testable service for order recovery:
  • fetchCoordinatorOrders(apiBase, { ethAddress, stellarAddress }) — issues one correctly-shaped request per connected address (since the coordinator only supports a single address filter), tolerates one side failing, and only throws (triggering the local-cache fallback) when every request fails.
  • mergeTransactions(local, remote) — de-dupes local and recovered orders by hashlock / on-chain order id / tx hash (falling back to id), letting the coordinator's authoritative record win when both sides describe the same order.
  • mapCoordinatorOrderToTransaction, isRealHash, isRealTransaction — moved out of the component so they're independently testable, with hashlock now carried through the mapping.
• frontend/src/components/TransactionHistory.tsx — wired to the fixed service; refreshFromCoordinator now uses the corrected request shape and mergeTransactions instead of a naive id-only Map.
• RefundDialog — no changes needed; its refund-eligibility gating already reads onChainOrderId / htlcContractAddress / timelockUnixSeconds off the mapped Transaction, which recovered orders now populate correctly since the request actually succeeds.

Tests added

• frontend/src/lib/orderRecovery.test.ts (17 tests): mapping coordinator orders → UI transactions, dedup rules across hashlock/order id/tx hash/id, per-address fetch fanout, partial vs. total coordinator failure, fake/demo hash filtering.
• frontend/src/components/TransactionHistory.test.tsx (4 tests): recovery renders after reconnecting a wallet, duplicate suppression between a locally-pending order and its coordinator-recovered counterpart, fallback to the local cache when the coordinator is unreachable, and no coordinator call when no wallet is connected.

Verification

• tsc --noEmit — clean
• vite build — succeeds
• vitest run (frontend) — all new tests pass (21/21); pre-existing failures in App.test.tsx (5, DeploymentSelfCheck mock issue) are unrelated and present on master prior to this change (verified via git stash).
• vitest run (coordinator) — http-routes.test.ts (20 tests) passes unmodified; no backend changes were required.

Closes #415

[vercel]

@dykdee is attempting to deploy a commit to the karagoz's projects Team on Vercel.

A member of the Team first needs to authorize it.

[dykdee]

Hello repo maintainer, please can you approve the workflows quickly? the issue is fixed but you have not attended to the case.

[dykdee]

Hi repo maintainer, please can you check that this issue is resolved and close it ASAP.