Add CodSpeed continuous performance measurement

[codspeed-hq]

Summary

This PR sets up CodSpeed for continuous performance measurement of River's Rust core (the stats module in rust_src, exposed to Python via PyO3).

River already ships a set of criterion benchmarks in benches/stats_bench.rs covering the streaming statistics primitives (mean, variance, EW mean/variance, kurtosis, skew, quantile, rolling quantile, rolling IQR). This change wires those benchmarks into CodSpeed so performance is tracked automatically on every pull request and on pushes to main.

Changes

• Cargo.toml: Swapped the criterion dev-dependency for codspeed-criterion-compat (aliased as criterion, version 5.0.1). The compat layer is a drop-in replacement, so benches/stats_bench.rs is unchanged. Also declared the stats_bench bench target with harness = false so it is picked up by cargo codspeed.
• Cargo.lock: Updated to reflect the new dependency.
• .github/workflows/codspeed.yml: New workflow that builds and runs the benchmarks under CodSpeed. It uses OIDC authentication (no token secret required) and runs all three requested instruments via a matrix:
  • CPU Simulation (instrumentation) on ubuntu-latest — deterministic, low-variance CPU measurements with flamegraphs.
  • Memory (memory) on ubuntu-latest — heap allocation and peak-memory tracking.
  • Wall time (walltime) on codspeed-macro runners — real-world elapsed time on consistent bare-metal hardware.
• README.md: Added the CodSpeed badge alongside the existing status badges.

Verification

The setup was validated locally with the CodSpeed CLI:

• cargo codspeed build and codspeed run --mode instrumentation -- cargo codspeed run completed successfully and reported all 13 benchmarks.
• cargo codspeed build -m memory and cargo codspeed build -m walltime both compile cleanly. Their runtime measurement could not be exercised in the local sandbox (memory needs an eBPF/BTF-capable kernel and walltime needs perf), but both run on CodSpeed's CI runners.

Next steps

• Merge to start collecting a performance baseline on main.
• Subsequent PRs will automatically get a performance report comparing against the base branch.
• Consider extending coverage to other hot paths as more of River's core moves to Rust.

[codspeed-hq]

Congrats! CodSpeed is installed 🎉

🆕 26 new benchmarks were detected.

You will start to see performance impacts in the reports once the benchmarks are run from your default branch.

Detected benchmarks

• ewmean_10k (Simulation): 63.8 µs
• ewvariance_10k (Simulation): 74.2 µs
• kurtosis_10k (Simulation): 175.8 µs
• mean_10k (Simulation): 74.2 µs
• quantile_p2_10k (Simulation): 499.9 µs
• rolling_iqr[10] (Simulation): 1.4 ms
• rolling_iqr[100] (Simulation): 1.9 ms
• rolling_iqr[1000] (Simulation): 3.7 ms
• rolling_quantile[10] (Simulation): 1.4 ms
• rolling_quantile[100] (Simulation): 1.9 ms
• rolling_quantile[1000] (Simulation): 3.7 ms
• skew_10k (Simulation): 120.1 µs
• variance_10k (Simulation): 85.4 µs
• ewmean_10k (Memory): 0 B
• ewvariance_10k (Memory): 0 B
• kurtosis_10k (Memory): 0 B
• mean_10k (Memory): 0 B
• quantile_p2_10k (Memory): 0 B
• rolling_iqr[10] (Memory): 160 B
• rolling_iqr[100] (Memory): 1.6 KB
• ...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

---