Fix silent RPC stalls: deadlines, diagnostics, PAPI 3.1

[kogeler]

Summary

• --validator-info no longer scans the entire Staking.Ledger map. It resolves
  each configured stash through Staking.Bonded plus a direct Ledger lookup,
  which is what previously made the mode hang on public Asset Hub endpoints. Chain
  work for 7 validators now takes ~8s instead of stalling indefinitely.
• Every chain call runs behind a per-call deadline, a retry budget with exponential
  backoff, and a global concurrency semaphore (default 16). The mode previously
  fired ~1150 simultaneous requests over a single WebSocket, which silently stalled
  most endpoints. All four limits are configurable in config.json.
• Failures are now reported instead of swallowed: a startup probe fails fast on an
  endpoint that accepts the socket but never answers, socket transitions are logged,
  stderr lines carry timestamps and per-phase durations, and the process always
  prints an EXIT code N line. Anything explaining termination is written with
  fs.writeSync, because console.error is asynchronous on a pipe and drops its
  buffer when process.exit follows in the same tick — that defect previously
  turned a real error into a bare exit code 1.
• Credentials embedded in rpcUrl are redacted in logs; the endpoint API key was
  previously printed to stderr in cleartext.
• Upgraded polkadot-api 2.2.2 → 3.1.0 and replaced the now-unreachable default and
  test-fixture RPC endpoint. The 3.0 breaking changes only affect transaction
  submission, which this read-only tool does not use; report output was verified
  byte-identical across both versions.