apollo_starknet_client: add request and connect timeouts to the http client

[matanl-starkware]

Why

RCA of the Testnet node-3 "Consensus Height Diff From Sync" spike to ~335 on Aug 12 (00:24–00:33 UTC): a single get_block?blockNumber=latest request from central sync got wedged inside the GCLB front end for 612s (logged as 502 internal_error with no serverIp, 20× the backend's 30s timeoutSec). The reqwest client is built with no timeout at all, so the call hung until central sync's no-progress watchdog fired on its 5-minute granularity — a 9m17s sync stall (~335 blocks) from one dead request.

The retry layer already classifies reqwest timeout errors as retryable (RetryErrorCode::Timeout), but reqwest never produces one without a configured timeout, so it could never engage.

What

Add a 60s total request timeout and a 10s connect timeout to the shared StarknetClient reqwest client (feeder + gateway clients). A hung connection now surfaces as a retryable timeout and request_with_retry re-issues the request on a fresh connection, bounding the stall at seconds instead of ~10 minutes. 60s is far above feeder p99 even for multi-MB class-definition responses.

Verification

• cargo build -p apollo_starknet_client — clean
• SEED=0 cargo test -p apollo_starknet_client — 76 passed, 0 failed
• cargo clippy -p apollo_starknet_client --all-targets --all-features — clean
• scripts/rust_fmt.sh — clean

🤖 Generated with Claude Code

[cursor]

PR Summary

Medium Risk
Touches all feeder/gateway HTTP traffic; legitimately slow responses could now time out and trigger retries, though 60s is intended to stay above normal p99 latency.

Overview
Configures the shared StarknetClient reqwest builder with a 60s overall request timeout and 10s connect timeout (new REQUEST_TIMEOUT / CONNECT_TIMEOUT constants), so hung feeder/gateway calls fail instead of blocking callers indefinitely.

That lets existing request_with_retry logic treat reqwest timeout errors as RetryErrorCode::Timeout and retry on a fresh connection, capping stalls to seconds rather than multi-minute sync watchdog delays.

^{Reviewed by Cursor Bugbot for commit fb04b25. Bugbot is set up for automated code reviews on this repo. Configure here.}

[reviewable-StarkWare]

This change is 

[ron-starkware]

@ron-starkware reviewed 1 file and all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on matanl-starkware).

[gkaempfer]

Security scan complete — no issues detected.

---

Generated by Claude Code