[codex] Optimize primitive arg layout fast paths

[MilkBlock]

This PR combines two tightly-related pieces needed to make the primitive-argument layout optimization meaningful on real workloads.

1. A narrow runtime fast path in core-relations for the single-External shape: when an action batch executes exactly one external call whose arguments are a consecutive variable window in the current struct-of-arrays bindings layout, the runtime batch-transposes once into a row-major scratch buffer and then invokes the external function row-by-row from that transposed block.
2. A rule-local query-entry normalization in core-relations::RuleBuilder::build_with_description(): for eligible single-external rules, external argument vars are normalized into a consecutive core-variable window while preserving their semantic order. This increases the hit rate of the runtime fast path in realistic workloads instead of relying on lucky pre-existing variable numbering.

The optimization is still intentionally scoped:

• no new public API
• no ruleset-wide scheduling changes
• no generalized argument-layout framework
• fallback remains the old path for any rule shape that does not match the narrow optimized corridor

Focused validation performed:

• cargo test -q -p egglog-core-relations run_instrs_single_external_borrowed_window_executes -- --nocapture
• cargo test -q -p egglog-core-relations single_external_rule_normalizes_gap_vars_into_consecutive_window -- --nocapture
• cargo test -q -p egglog-bridge constrain_prims -- --nocapture
• cargo check -q

Downstream workload validation:

• I compared eggplant math_microbenchmark before/after on the corresponding downstream dependency setup.
• With the finer-grained internal breakdown enabled, the relevant stage improved as follows:
  • rewrites run_ruleset: 1.605385708 s -> 1.503743250 s (~6.3% faster)
  • total benchmark time: 3.005323708 s -> 2.813726958 s (~6.4% faster)

That is the reason these two changes are folded together here: the runtime transpose lane alone is too narrow to justify a broader performance claim, while the query-entry normalization makes the same lane actually trigger in the real single-external rule shapes we care about.

[codspeed-hq]

Merging this PR will not alter performance

✅ 38 untouched benchmarks
⏩ 190 skipped benchmarks¹

---

_{Comparing MilkBlock:codex/rust-rule-layout-fastpath-pr (ef04f91) with main (21b99b6)}

Footnotes

1. 190 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 93.69369% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 86.29%. Comparing base (912663b) to head (ef04f91).
⚠️ Report is 110 commits behind head on main.

Files with missing lines	Patch %	Lines
core-relations/src/query.rs	92.75%	5 Missing ⚠️
core-relations/src/action/mod.rs	95.23%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #845      +/-   ##
==========================================
+ Coverage   81.54%   86.29%   +4.74%     
==========================================
  Files          88       85       -3     
  Lines       24112    23373     -739     
==========================================
+ Hits        19662    20169     +507     
+ Misses       4450     3204    -1246     

☔ View full report in Codecov by Sentry.
📢 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.