ci(bench): do not fail benchmarks when Polar Signals setup fails

[joseph-isaacs]

Rationale for this change

The Setup Polar Signals step downloads the Parca Agent release binary over the network, and that download can fail transiently. In run 31181958893 it failed with socket hang up three times in a ~28s window (the action's own retries at 18s and 10s), which aborted the compression-bench job before the benchmark ran.

That failure was a flake, not a config problem: 15 other jobs in the same run executed the identical step — same pinned action, same parca_agent_version: "0.49.0" — and succeeded, two of them within seconds of the failure.

Profiling is supplementary to the benchmark results, so a dropped connection while fetching the agent should not cost us the whole benchmark job. The adjacent Pre-upload benchmark debuginfo to Polar Signals step is already continue-on-error: true; this makes the setup step consistent with it.

What changes are included in this PR?

Adds continue-on-error: true (with a one-line comment explaining why) to the Setup Polar Signals step in the three workflows that run it:

• .github/workflows/pr-bench-runner.yml
• .github/workflows/develop-bench.yml
• .github/workflows/sql-bench-matrix.yml

When the step fails, the job now continues into Setup benchmark environment and runs the benchmark without profiling data, instead of failing outright. The step is still reported as failed in the job summary, so the flake stays visible.

Checks run: yamllint --strict -c .yamllint.yaml on all three changed workflows. No Rust or other code is touched, so no cargo checks were run.

What APIs are changed? Are there any user-facing changes?

None. CI configuration only.

---

Generated by Claude Code

[codspeed-hq]

Merging this PR will degrade performance by 0.11%

⚡ 2 improved benchmarks
❌ 2 regressed benchmarks
✅ 1933 untouched benchmarks
⏩ 51 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	decode_varbin[(1000, 2)]	62.1 µs	78.6 µs	-21.03%
❌	Simulation	decompress[u64, (1000, 16)]	65.1 µs	73.1 µs	-10.92%
⚡	Simulation	decompress[u32, (10000, 256)]	100.4 µs	83.6 µs	+20.11%
⚡	Simulation	cold_misaligned[(64, 256)]	5.1 ms	4.4 ms	+17.85%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing claude/parca-agent-download-failure-gha9fg (9c75ddd) with develop (66c447e)}

Footnotes

1. 51 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩