fix: benchmark comparison workflow suffix naming

[0xShaito]

Summary

• Fix benchmark comparison workflow showing Base: 0 by correcting suffix naming mismatch
• Add explicit benchmark execution step in PR workflow
• Use only_report: true flag to skip external action's unreliable benchmark execution

Root Cause

The comparison was failing because:

1. update-baseline.yml created files with _latest suffix
2. PR workflow expected to download baseline and compare with _new suffix
3. External action's implicit benchmark execution via npx aztec-benchmark was unreliable
4. No _new files were generated → comparison returned empty/zero results

Changes

.github/actions/benchmark/action.yml:

• Add explicit Run benchmarks step: yarn aztec-benchmark --suffix _latest
• Add reports_dir: ${{ inputs.bench-dir }}
• Change base_suffix from '_latest' to '_baseline'
• Change current_suffix from '_new' to '_latest'
• Add only_report: true to use external action only for comparison, not execution
• Update file cleanup/upload paths to use _baseline suffix

.github/workflows/update-baseline.yml:

• Change benchmark suffix from _latest to _baseline
• Update file listing and upload paths to use _baseline suffix

New Flow

update-baseline.yml (on push to dev/main):

1. Runs yarn aztec-benchmark --suffix _baseline
2. Uploads *_baseline.benchmark.json as artifact

benchmark/action.yml (on PR):

1. Downloads baseline artifact (*_baseline.benchmark.json)
2. Runs yarn aztec-benchmark --suffix _latest (generates current)
3. External action compares _baseline vs _latest (only_report: true)
4. Posts comparison, renames files to _baseline, uploads for PR branch

Test plan

• After merge, update-baseline.yml workflow runs and creates baseline with _baseline suffix
• Future PRs correctly download baseline and generate comparison reports with proper values

🤖 Generated with Claude Code

[github-actions]

Benchmark Comparison

CPU	Cores	RAM	Arch
N/A	N/A	N/A	N/A

Contract: counter

	Function	Gates			DA Gas			L2 Gas			Proving Time (ms)
Status		Base	PR	Diff	Base	PR	Diff	Base	PR	Diff	Base	PR	Diff
🆕	increment	0	472,292	+Inf%	0	2,253	+Inf%	0	28,872	+Inf%	N/A	N/A

[0xShaito]

Closing - reverting to original naming on test/benchmark instead.