Fix 429 rate limits — batch transport, memory cache, more endpoints

[realproject7]

Summary

• Batch transport: Enable batch: true on both browserClient and publicClient — viem combines multiple pending readContract() calls into a single JSON-RPC batch HTTP request
• Faster failover: Reduce timeout from 5-10s to 2s on all transports so endpoints rotate sooner on failure
• 12 RPC endpoints: Expand CORS and server endpoint lists from 5/8 to 12 each (matching mintpad/mint.club patterns)
• Singleton memory cache (lib/cache.ts): 60s TTL + in-flight request deduplication — prevents duplicate concurrent RPC calls for the same token
• Price function caching: getTokenPrice(), get24hPriceChange(), getTokenTVL(), getBatchTokenData() all wrapped with priceCache.get()

Expected Impact

Before	After
15+ HTTP requests per page load	2-3 batched HTTP requests
5 CORS endpoints, 5-10s timeout	12 endpoints, 2s timeout
No dedup, duplicate concurrent calls	Singleton cache deduplicates
429 errors in console	Clean console

Files Changed

File	Change
lib/rpc.ts	batch: true, 2s timeout, 12 endpoints, new display names
lib/cache.ts	New — singleton MemoryCache with TTL + in-flight dedup
lib/price.ts	All 4 price functions wrapped with priceCache.get()

Test plan

• npm run typecheck passes
• npm run lint passes
• Browse homepage — no 429 errors in console
• Navigate to token page — price loads from cache on repeat visits within 60s
• Network tab shows batched JSON-RPC requests (multiple calls in single HTTP body)

Fixes #399

🤖 Generated with Claude Code

[project7-interns]

Verdict: APPROVE

Summary

The PR matches issue #399 with the intended narrow scope: batched RPC transports, expanded endpoint pools, and shared in-memory dedup caching for the targeted price helpers. I did not find a correctness or design regression in the changed paths, and the required lint/typecheck job is passing.

Findings

• None.

Decision

Approve: the change satisfies the acceptance criteria and keeps the fix isolated to the requested lib/ surface area.

[project7-interns]

Both issues resolved cleanly. Errors now propagate through the cache (no stale null caching), and custom-client calls bypass the cache entirely. LGTM.