fix(p2p): reduce flakiness in proposal tx collector benchmark

[AztecBot]

Summary

Fixes flakiness in p2p_client.proposal_tx_collector.bench.test.ts caused by three compounding issues:

1. chunkTxHashesRequest defaulted to chunkSize=1, creating 500 individual libp2p streams for the 500-tx send-batch-request case. The rapid stream churn overwhelms the connection, causing EPIPE cascades that kill the muxer. Bumped to chunkSize=8 as the existing TODO indicated.

2. Peer scores persisted between benchmark cases, so hundreds of HighToleranceError penalties from EPIPE failures in one case degraded peer selection in subsequent cases. Added PeerScoring.resetAllScores() and called it in the worker before each benchmark run.

3. No connectivity check between cases, so degraded connections from a previous case could silently affect the next. Added waitForConnectivity() to verify the aggregator has 80% of expected peers before each case starts.

Full analysis with CI log evidence: https://gist.github.com/AztecBot/e5af3238fbfefc29c51de2ee5deaa8ea

Changes

• protocols/tx.ts: Change chunkTxHashesRequest default chunkSize from 1 to 8
• peer_scoring.ts: Add resetAllScores() method
• p2p_client_testbench_worker.ts: Reset peer scores before each bench case, add GET_PEER_COUNT IPC command
• worker_client_manager.ts: Add waitForConnectivity() and getPeerCount() methods
• p2p_client.proposal_tx_collector.bench.test.ts: Check connectivity in beforeEach

ClaudeBox log: https://claudebox.work/s/38590d3cfe6a7000?run=2

[AztecBot]

Automatically closing this stale claudebox draft PR (no updates for 5+ days). Re-open if still needed.