fix(builder): bound Flashbots relay requests with a 30s client timeout

[hermiondev]

Problem

FlashbotsClient::new builds its HTTP client with reqwest::Client::new(), whose default is no timeout on requests. If the Flashbots relay (or a configured builder relay) accepts the TCP connection but stalls — overloaded relay, middlebox blackhole, hung LB — send_private_transaction and cancel_private_transaction never settle:

• the sender task's submit/poll loop is pinned forever on one socket,
• bundle submission for the whole builder halts silently (no error, no retry, no log),
• cancels never land, leaving stuck transactions un-replaced.

Every other outbound HTTP path in this repo is already bounded — the provider uses a tower timeout layer (#1297, client_timeout_seconds), the healthcheck RPC calls use rpc_timeout (#1173), the tracer uses 10s (#730). The Flashbots sender was the last unbounded Client::new() in production code (grep -rn "Client::new()" crates/ → this line only).

Fix

Client::builder().timeout(Duration::from_secs(30)) on the Flashbots relay client. 30s matches the order of the other concrete bounds in the repo and is generous for eth_sendPrivateTransaction / cancel round-trips. On timeout reqwest returns a RequestTimedOut transport error, which the existing FlashbotsError::Transport arm already maps → normal retry/cancel semantics resume instead of a hang.

Verification

• cargo check -p rundler-builder — Finished, 0 errors (rustc 1.92.0, repo pinned toolchain, request resolved from Cargo.lock reqwest 0.12.9).
• cargo fmt -p rundler-builder -- --check — clean.
• Single-file change, +8/−1, no behavior change on the success path.