fix(cypher): apply distinct before return limits

[pcristin]

Summary

• Apply RETURN DISTINCT before ORDER BY, SKIP, and LIMIT post-processing.
• Preserve the simple RETURN ... LIMIT fast path only for non-DISTINCT, non-ordered returns.
• Add regressions for RETURN DISTINCT with ORDER BY/LIMIT, LIMIT only, and ORDER BY/SKIP/LIMIT.

Root Cause

RETURN rows were sorted and truncated before de-duplication. When duplicate projected rows appeared before the limit boundary, those duplicates consumed the limit and the final DISTINCT pass returned too few rows.

Test Plan

• make -f Makefile.cbm build/c/test-runner
• ./build/c/test-runner 2>&1 | awk '/=== cypher ===/{show=1} show{print} /cypher_parse_unwind_var/{exit}'
  • Passed the Cypher regression window, including the new issue 873 tests.
• git diff --check origin/main..HEAD
• scripts/test.sh
  • Failed in the existing incremental/MCP tooling suite with repeated pipeline.err phase=dump failures and an AddressSanitizer leak summary.
  • The observed failures were outside the Cypher query execution path changed here.
• scripts/lint.sh
  • Failed after installing clang-tidy, clang-format, and cppcheck.
  • clang-tidy reports repository-wide warnings-as-errors outside this change.
  • cppcheck reports src/pipeline/pass_pkgmap.c:421 null-pointer warning and an internal AST error at internal/cbm/extract_defs.c:3386.

Notes

Fixes #873
Refs #594

[DeusData]

Thanks for the #873 Cypher fix. Triage: high-priority query correctness bug.

Review focus: DISTINCT must happen before ORDER BY/SKIP/LIMIT for the affected shapes, while the non-DISTINCT fast path stays intact. The reported unrelated suite failures should stay clearly separated from this diff during review.