Fix relationship-row aggregate multiplicity for Cypher collect/count/sum (#1343)

[lmeyerov]

Summary

Fixes #1343 by routing multiplicity-sensitive Cypher aggregates on relationship-pattern MATCH queries through bindings-row execution so row multiplicity is preserved before aggregation.

What changed

• Added aggregate-shape detection in lowering for relationship-pattern MATCH + multiplicity-sensitive aggregate usage.
• For those shapes, force bindings-row path (rows(binding_ops=...)) so repeated relationship rows survive to group_by.
• Scoped the old repeated MATCH rows aggregate guard to non-bindings paths.
• Added admission constraints so forced bindings routing is limited to safe lanes (preserving existing fail-fast behavior for known unsupported multi-source/whole-row overlap lanes).
• Added regression coverage for:
  • grouped/global relationship-row count(*), count(alias), sum(1), avg(edge_prop)
  • OPTIONAL MATCH ... WITH p, collect(uni.name)
  • OPTIONAL MATCH ... WITH p, collect(CASE ...)
  • grouped relationship-row WITH a, count/sum/avg(...)
• Added changelog entry in Development.

CI guard update

• Updated bin/ci_cypher_surface_guard_baseline.json to match intentional lowering.py growth from this feature and follow-up safety constraints.
• Verified guard locally with python bin/ci_cypher_surface_guard.py (pass).

Validation

• ./bin/ruff.sh graphistry/compute/gfql/cypher/lowering.py graphistry/tests/compute/gfql/cypher/test_lowering.py
• python -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py
• python -m pytest -q tests/gfql/ref/test_df_executor_core.py -k "missing_alias_raises or missing_where_column_raises_during_input_build or where_missing_column_after_prior_call_still_rejected"
• python bin/ci_cypher_surface_guard.py

Notes

• This PR focuses on relationship-row multiplicity for aggregate expressions over scalar/property projections.
• Whole-row carrier grouping shape (RETURN a, count(*)) remains on the legacy representation path and is not expanded in this change.