Ci benchmarks (#2019)

* feat: CI benchmark regression detection

Add a consolidated benchmark suite (`ci_benchmark`) and a GitHub Actions
workflow that automatically compares performance against the main branch
baseline on every PR.

Benchmark coverage (13 measurements, ~4 min on CI):
  - BPE GPT-2: encode, batch, no-cache, batch-no-cache
  - Llama-3: encode, batch, encode-fast, char-offsets, concurrent-4t
  - Serialization: GPT-2 load, Llama-3 load, Llama-3 save
  - Training: BPE small corpus

CI workflow:
  - On push to main: run benchmarks, store baseline in gh-pages branch
  - On PR: run benchmarks, compare vs baseline, post/update a single
    PR comment with the delta table
  - Alert threshold: 15% regression (warn, don't fail)

Uses benchmark-action/github-action-benchmark with criterion's bencher
output format for machine-readable results.

* fix: rustfmt + add github-token for PR comments

* ci: require approval for benchmark runs on PRs

* bench: add from_file + deserialize benchmarks to ci_benchmark

* fix: split workflow into two jobs to avoid empty environment value

* fix: download benchmark data in CI + skip if output empty

* fix: use huggingface-cli for data download (handles gated models like Llama-3)

* fix: use uvx for huggingface-cli, add setup-uv step

* fix: hf not huggingface-cli

* fix: single download from hf-internal-testing/tokenizers-bench-data

* fix: use hf-internal-testing dataset for BOTH jobs, remove all curl/gated refs

* Initialize gh-pages for benchmark data

* upupdate

* Initialize gh-pages for benchmark data

* ci: sccache, workflow_dispatch with PR comment, push-to-main only (no PR trigger)

* ci: temporarily trigger on ci-benchmarks branch for testing

* ci: touch tokenizers/ to trigger path filter

* ci: store baselines on HF Hub, drop github-action-benchmark + gh-pages

* ci: also trigger on workflow file changes + doc touch

* ci: skip upload if HF_TOKEN missing, remove ci-benchmarks branch trigger

* feat: add Python benchmark suite + CI matrix for Python bindings perf

- pytest-benchmark based test suite covering:
  - BPE GPT-2: encode, encode_batch, multithreaded (4 workers)
  - Llama-3: encode, encode_batch, encode_fast, multithreaded, decode_batch
  - Async: async_encode_batch, async_encode_batch_fast
  - Serialization: from_file, to_str, from_str (roberta, llama3, albert)
  - Training: BPE small corpus
- CI workflow: separate benchmark-python job with sccache + maturin
- Re-enabled ci-benchmarks branch trigger for testing

* fix: skip benchmark tests when pytest-benchmark is not installed

* ci: Python bench comparison against saved baseline + PR comment

* ci: compare against baseline BEFORE uploading new one (both Rust and Python)

* ci: /benchmark comment trigger + only upload baseline on push to main

- New benchmark-trigger.yml: maintainer comments '/benchmark' on a PR
  to dispatch the benchmark workflow on the PR's ref
- Upload steps gated on github.event_name == 'push' so workflow_dispatch
  (PR runs) never overwrite the baseline
- Trigger requires MEMBER/OWNER/COLLABORATOR association

* ci: remove ci-benchmarks branch trigger

* ci: /benchmark creates a check run on the PR, updated with results when done

- benchmark-trigger.yml creates a 'Benchmark Results' check on the PR head SHA
- benchmarks.yml marks it in_progress at start, completed (success/failure) at end
- The check body contains the comparison markdown table
- Can be made a required check in branch protection rules

* ci: use criterion --save-baseline/--baseline + export artifacts

Rust:
- Push to main: runs with --save-baseline main, uploads criterion
  data (tar.gz) + bencher output to HF Hub
- workflow_dispatch: downloads criterion baseline, runs with --baseline main
  for automatic criterion comparison
- criterion HTML report uploaded as GitHub Actions artifact (30 day retention)

Python:
- bench_output.json uploaded as GitHub Actions artifact
- Baseline stored/compared via HF Hub as before

Both:
- Artifacts downloadable from the workflow run page for manual inspection
- Comparison tables posted to PR comments

* update

* ci: pin macOS Python to 3.13 (3.14 breaks abi3 linking)

* ci: add -undefined dynamic_lookup for macOS abi3 cross-compilation linking

* ?

* ci: ubuntu-latest-4-cores + SVG benchmark charts with red/green color coding

- Switch both jobs to ubuntu-latest-4-cores for more consistent results
- Replace plaintext markdown tables with SVG charts generated by
  .github/scripts/render_bench_svg.py
- Dark theme, monospace font, red/green bars + delta percentages
- SVGs uploaded as artifacts and embedded in PR comments
- Supports both Rust (bencher format) and Python (pytest-benchmark JSON)

* Apply suggestion from @ArthurZucker

* Apply suggestions from code review

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* fix: render bench charts as PNG for GitHub comment compatibility

SVG doesn't render in GitHub PR comments. Now:
- render_bench_svg.py supports --output .png via cairosvg
- PNGs uploaded to hf-internal-testing/tokenizers-bench/charts/
- PR comments embed as ![img](hf_url) which GitHub renders correctly

* ci: bump bench rounds to 15 (Rust sample_size + Python min-rounds)

* fix: use uv pip install --system for cairosvg