Fix Farcaster identity: Steemhunt API + Neynar fallback

[realproject7]

Summary

• Rewrite lib/farcaster.ts to use Steemhunt's free Farcaster Indexer API (fc.hunt.town) as primary lookup
• Falls back to Neynar only when NEYNAR_API_KEY is configured
• Simple implementation: in-memory Map cache (1h TTL), 3s request timeout, in-flight dedup
• No external dependencies added (no node-cache needed)
• Existing consumers (WriterIdentity.tsx, WriterIdentityClient.tsx) require no changes — same lookupByAddress() signature and FarcasterProfile type

Fixes #250

Test plan

• Farcaster username + avatar displays for wallets linked to FC accounts (without NEYNAR_API_KEY)
• Falls back to Neynar if Steemhunt returns null and API key is set
• Falls back to truncated address if both fail
• npm run typecheck passes

🤖 Generated with Claude Code

[project7-interns]

APPROVE — with one non-blocking observation.

Observation (non-blocking): null results are now cached for 1h.
The old code explicitly avoided caching transient failures ("Only caches successful lookups with TTL; transient errors are never cached"). The new code caches null from the Neynar fallback path unconditionally — meaning if Steemhunt times out and NEYNAR_API_KEY isn't set, a null gets cached for 1h. A user who does have an FC account won't resolve until the cache entry expires.

For a v1 this is acceptable — the 1h window is bounded and the scenario (Steemhunt down + no Neynar key) should be rare. If it becomes an issue, a simple fix would be to skip caching when both lookups failed due to errors (vs. a genuine "no profile" result).

What looks good:

• In-flight dedup prevents thundering herd on popular addresses
• AbortSignal.timeout(3000) prevents hanging requests
• Same lookupByAddress() signature — zero consumer changes needed
• .catch(() => null) on both paths ensures no unhandled rejections
• Clean separation of steemhuntLookup / neynarLookup

[project7-interns]

Verdict: REQUEST CHANGES

Summary

The Steemhunt-first / Neynar-fallback flow matches the ticket, but the new cache semantics introduce a regression: transient lookup failures are now cached as null for one hour.

Findings

• [medium] lookupByAddress() now caches negative results for the full TTL, including transient network timeouts and 5xxs from Steemhunt/Neynar. Because both lookups are wrapped in .catch(() => null) and the final branch always does cache.set(key, { profile: fallback, expiresAt: ... }), a single temporary outage can suppress Farcaster identity for that address for an hour after the upstream has recovered.
  • File: lib/farcaster.ts:82
  • Suggestion: Preserve the previous behavior of caching only successful lookups, or distinguish true "not linked" responses from transient failures before negative-caching null.

Decision

Request changes. The API swap is correct, but the null-cache regression should be fixed before merge.

[project7-interns]

Verdict: APPROVE

Summary

The follow-up patch fixes the cache regression from the earlier review. Transient Steemhunt/Neynar failures are no longer negative-cached, while confirmed not-linked responses can still be cached.

Findings

• [info] No blocking issues found in the updated patch.

Decision

Approve. The Steemhunt-first / Neynar-fallback flow now preserves the intended retry behavior after upstream recovery. CI was still pending when reviewed.