Optimize repeated queryplan term analysis

[carli2]

What

• replace deep equal? dedup in hot scalar/domain helper paths with serialized string keys
• pre-group aggregate scalar parts by alias instead of rescanning the same join-condition list per alias
• memoize top-level pure AST analysis helpers for repeated planner traversals:
  • extract_tblvars
  • extract_columns_for_tblvar
  • extract_outer_columns_for_tblvar
  • unnest_expr_outer_refs

Why

Large app queries compile through repeated planner traversals over the same nested scalar and permission terms. This PR reduces repeated term analysis without changing lowering semantics.

Validation

• python3 run_sql_tests.py tests/32_expr_subselects.yaml
• python3 run_sql_tests.py tests/117_dashboard_unnesting_regressions.yaml
• make test was started locally and ran through the broad suite without failures in the observed output; CI will provide the definitive full run

Cold benchmark

Fresh restart / fresh in-memory queryplan cache, same copied app data and badge query:

• master: 28.78s
• this branch: 29.09s

So the branch is currently correctness-safe but benchmark-neutral on that exact homepage badge query. I am still opening it because the optimization direction is general and the changes are isolated enough for CI review.

[carli2]

nonsense