fix: restore 255 missing benchmarks — jest-mock-extended crash + parallelize_strict error swallowing

[AztecBot]

Summary

Since 2026-03-18 (PR #21627), 255 benchmark metrics stopped being published to the benchmark dashboard. This includes bb-micro-bench, l1-contracts gas, sol verifier gas, noir circuit gate/opcode counts, p2p benchmarks, prover-client benchmarks, and AVM test benchmarks.

Root Cause

Two interacting bugs:

1. jest-mock-extended crash in bench workers: PR fix: (A-649) tx collector bench test #21619 changed p2p_client_testbench_worker.ts and proposal_tx_collector_worker.ts to use mock<EpochCache>() from jest-mock-extended. These files run as Node worker processes outside Jest, so the import crashes immediately.

2. parallelize_strict silently swallows failures: run_tests | tee $output puts run_tests in a subshell (left side of pipe). When the p2p bench fails, wait_for_job calls exit 1 — but this only exits the subshell. The main script sees no background jobs (they were children of the subshell) and exits 0. This prevents ALL subsequent benchmarks from being scheduled, while silently reporting success.

Fix

• Replace mock<EpochCache>() with createMockEpochCache() in both bench worker files — this helper is already defined in testbench-utils.ts and has no Jest dependency
• Fix parallelize_strict to use process substitution (> >(tee $output)) instead of a pipe, keeping run_tests in the main shell so background jobs remain trackable
• Change wait_for_job to set a failed flag instead of exit 1, allowing remaining benchmarks to continue running
• Update bench_engine to tolerate failures from parallelize_strict so partial results are still collected and uploaded

ClaudeBox log: https://claudebox.work/s/f0f8be7210f49dcb?run=2