fix(fx): keep exchange rates warm and decouple the failure cooldown from the timeout

[zaxovaiko]

Summary

Keeps configured exchange rates warm on a 30s schedule, gives the failure cooldown its own exchangeRate.failureCooldownMs knob (default 30s), and stops useExchangeRate(s) refetching inside the fresh window. BF-0

Why

• Nothing refreshed rates on a schedule. Once a rate went hard-stale, the next caller fetched it synchronously, including callers holding a locked transaction (RG limit gate, rank accrual, social transfers), so they waited up to providerTimeoutMs on the vendor.
• After a failed fetch the cooldown was exactly providerTimeoutMs, so during an outage almost every call waited out the timeout again.
• The react hooks had no staleTime, so every mount and window refocus refetched a rate the server still considered fresh.

The warm-up reuses the existing getRate path and job-queue pattern (same shape as the wallet sweep cron): every wallet.cryptoCurrencies entry plus display currencies, minus the pivot, bounded at 4 concurrent fetches, per-currency failures logged and swallowed.

Alternatives considered

• Longer hardMaxAgeMs only: hides the stall but serves older prices to limit checks.
• Moving conversions out of locked transactions: larger refactor of each caller; warm rates remove the synchronous fetch in practice.

Risks

• New config field with a default; no migration. Operators with a vendor rate limit see one fetch per currency per soft-stale window (~every 60-90s).
• A rate missing from the warm list still falls back to the on-demand fetch.