Gallop sorted free-join intersections

[ezrosent]

Summary

• Intersect sorted two-way generic-join probes with a two-finger merge for small or similarly sized inputs and exponential galloping for skewed inputs.
• Preserve row-carrying ProbeMatch semantics for Packed, ProjectedRoot, and SmallColumn indexes, with the existing scalar fallback for hash-backed and unsupported shapes.
• Drop the serial action-batch increase after it changed fixed-iteration eggcc-2mm results and regressed stresstest_large_expr.

Stack

Draft, stacked on #962. This PR now contains one focused performance commit.

Performance

Local wall-time comparisons against #962 after removing the batch-size change:

• stresstest_large_expr: 196.7 ms base vs 196.8 ms head median (neutral).
• eggcc-2mm: 965.5 ms base vs 949.7 ms head median (1.6% faster).
• Three-sample neural-network smoke checks ranged from about 4.1% faster on Gemma4-MoE to about 1.4% slower on Llama; the smaller movements are within the noise floor of this short run.

The previous CodSpeed PR comment compares the now-removed a9c79c8 batch-size version and also warns that its base and head used different runtime environments. GitHub did not emit a new pull_request:synchronize event for the stack force-update, so that bot comment remains stale.

The full Build workflow was explicitly dispatched on 3ae16b04; all 30 jobs passed, including fresh stresstest_large_expr and eggcc-2mm simulation shards:

https://github.com/egraphs-good/egglog/actions/runs/30564246486

Validation

• cargo fmt --check
• cargo clippy -p egglog-core-relations --all-targets -- -D warnings
• cargo test -p egglog-core-relations: 148 tests passed
• 2 doctests passed
• cargo test --release --test files -- eggcc_2mm: all three variants passed
• Full GitHub Build workflow: 30 of 30 jobs passed

[codspeed-hq]

Merging this PR will not alter performance

✅ 37 untouched benchmarks
⏩ 227 skipped benchmarks¹

---

_{Comparing codex/vectorized-free-join-on-962 (f93fb49) with ezr-better-rebuild (2f046f1)}

Footnotes

1. 227 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. ↩

[codecov-commenter]

Codecov Report

❌ Patch coverage is 94.42060% with 13 lines in your changes missing coverage. Please review.
✅ Project coverage is 87.27%. Comparing base (2f046f1) to head (f93fb49).

Files with missing lines	Patch %	Lines
core-relations/src/free_join/execute.rs	94.42%	13 Missing ⚠️

Additional details and impacted files

@@                  Coverage Diff                   @@
##           ezr-better-rebuild     #966      +/-   ##
======================================================
+ Coverage               87.21%   87.27%   +0.05%     
======================================================
  Files                      96       96              
  Lines                   34185    34407     +222     
======================================================
+ Hits                    29816    30030     +214     
- Misses                   4369     4377       +8     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.