RPC client retries permanent failures (DNS, auth errors)

Problem

All non-RPC errors trigger retries, but some network errors (DNS failures, invalid URLs) should not be retried.

Location

src/protocol/rpc.ts:102-117

Current Code

if (err instanceof RPCRequestError) {
  throw err;  // Don't retry RPC errors
}
// All other errors retry - including permanent failures
if (attempt < this.retries) {
  await sleep(this.retryDelay * (attempt + 1));
}

Recommendation

• Classify errors (transient vs permanent)
• Don't retry: DNS errors, invalid URL, auth failures
• Do retry: connection reset, timeout, 5xx errors

Priority

Critical - Correctness

Start at src/protocol/rpc.ts:102-117 and trace how errors reach the retry decision. Classify DNS failures, invalid URLs, and authentication failures as non-retryable, while preserving retries for connection resets, timeouts, and 5xx errors; done means each category follows the requested retry behavior.