feat: add per-method RPC timeout configuration

[KENILSHAHH]

Summary

• Previously all RPC calls shared no timeout, making it impossible to balance
  the tradeoff between witness fetches (which can legitimately take several
  seconds on heavy blocks) and block/header fetches (which should return in
  milliseconds). A single tight timeout caused false witness failures; a single
  loose one left workers hanging on stuck block fetches.
• Add block_timeout, witness_timeout, and code_timeout fields
  (Option<Duration>) to RpcClientConfig with corresponding with_* builder
  methods. All default to None — fully backwards compatible.
• Add a private apply_timeout helper on RpcClient that wraps any future with
  tokio::time::timeout when a timeout is set, and passes through with zero
  overhead when None.
• Apply code_timeout to get_code, block_timeout to get_block_unchecked /
  get_latest_block_number / get_header / get_transaction_by_hash, and
  witness_timeout to fetch_witness_from_provider (covers both primary and
  Cloudflare fallback paths).
• Expose --block-timeout-secs, --witness-timeout-secs, --code-timeout-secs
  CLI flags (and STATELESS_VALIDATOR_*_TIMEOUT_SECS env vars) in the
  stateless-validator binary so operators can tune timeouts at deploy time
  without recompiling.

[flyq]

Closing as superseded. The problem this addressed — RPC calls with no timeout, so a stalled provider could hang a worker — was fixed on main by #129: every provider attempt is bounded by per_attempt_timeout (default 20s, --rpc-per-attempt-timeout-ms, crates/stateless-common/src/rpc_client.rs:115), and an expired attempt rotates to the next provider instead of surfacing a timeout error to the caller. #172 added a 3s connect timeout so unreachable hosts fail fast, and #182 added a witness-specific cap for deadline-bound witness fetches. The files this PR touches have also moved (crates/stateless-common/src/rpc_client.rs, bin/stateless-validator/src/app.rs). Thanks for the contribution.