perf(sql): optimize expression evaluation

[ngjaying]

Summary

• Reuse the project evaluator and its valuer chain across input rows.
• Normalize numeric operands with one type switch instead of repeated type checks and cast dispatch.
• Compare common changed_cols value types directly before falling back to reflect.DeepEqual.

Why

• SQL expression evaluation runs for every input row, so small setup and dispatch costs accumulate in projection, arithmetic, filtering, sorting, and change detection.
• The project operator rebuilt the same evaluator scaffolding for every row even though only the current tuple, window, and aggregate data changed.
• Numeric normalization inspected each operand type multiple times, while changed_cols used reflection even for primitive values.

Changes In This PR

• Expose reusable internal multi-valuer implementations and reset their per-row data from ProjectOp without changing expression indices or SliceTuple output semantics.
• Replace convertNum helper dispatch with a single type switch that preserves the existing integer and floating-point conversions.
• Add primitive fast paths for string, int64, float64, bool, and int comparisons in changed_cols; complex values retain reflect.DeepEqual behavior.
• Add focused regression tests and benchmarks for all three paths.
• Organize the changes as three independent commits for review.

Notes

• Benchmark medians on linux/amd64, Intel i9-14900HX, seven runs:
  • int64 SimpleDataEval: 10.82 ns/op to 8.08 ns/op (about 25% faster), 0 allocations before and after.
  • SliceTuple projection: 158.6 ns/op to 54.9 ns/op (about 65% faster), from 112 B / 4 allocations to 0 B / 0 allocations.
  • stable changed_cols: 337.0 ns/op to 304.8 ns/op (about 10% faster), with allocations unchanged.
• Passed go test -race ./internal/xsql ./internal/topo/operator -count=1 and focused race tests for the changed column functions.
• git diff --check passes.
• A full local binder/function run still hits the pre-existing TestIncAggFunctionErr failure, reproduced unchanged on upstream/master. go vet likewise reports the existing lock-copy warning in untouched internal/xsql/row.go:187.

[codecov]

Codecov Report

❌ Patch coverage is 96.19048% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 70.78%. Comparing base (a1c06e8) to head (175ccd8).

Files with missing lines	Patch %	Lines
internal/xsql/valuer.go	91.84%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4075      +/-   ##
==========================================
+ Coverage   70.74%   70.78%   +0.03%     
==========================================
  Files         463      463              
  Lines       54283    54347      +64     
==========================================
+ Hits        38402    38466      +64     
- Misses      12888    12889       +1     
+ Partials     2993     2992       -1     

☔ 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.