feat(cypher): multi-alias scalar RETURN and edge alias properties (#981, #982) - #986

MATCH (a)-[:R]->(b) RETURN a.id AS a_id, b.id AS b_id now works.

Implementation: when multiple MATCH aliases appear in a scalar RETURN,
build a bindings table by joining edges with node properties for each
alias endpoint (src/dst). The joined table has columns prefixed with
alias names (a.id, b.id, etc.) that the existing row pipeline can
project from directly.

Key changes:
- lowering.py: _build_alias_endpoints() maps node aliases to src/dst
  edge endpoints; catch multi-alias rejection and use bindings path
- pipeline.py: rows(alias_endpoints=...) builds the bindings table
  via vectorized merge (pandas/cudf compatible)
- validation.py: allow alias_endpoints param on rows() safelist
- ast.py: rows() accepts optional alias_endpoints dict
- _ProjectionPlan: track all_source_aliases for multi-alias detection

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

…982)

Add 4 end-to-end tests for multi-alias RETURN:
- basic a.id + b.id projection
- multiple bindings (2 edges → 2 rows)
- empty match → empty result
- edge alias property access (r.creationDate) — validates #982 fix

DRY: remove duplicated projection/distinct/order_by/page logic from
try/except fallback path in _lower_projection_chain; both single and
multi-alias now share the same downstream code.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Only allow multi-alias RETURN for scalar projections (a.id, b.name).
Whole-row projections (RETURN n, x) with multiple aliases still raise
the unsupported error, as the bindings table approach doesn't handle
entity-valued outputs.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Only use bindings-table path for pure scalar alias.prop projections.
Reject multi-alias when whole-row outputs or expression-based outputs
(type(), count(), etc.) are present.

Fix mypy: use Optional[GFQLValidationError] for the caught exception.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Tighten multi-alias guard: only use bindings path when at least one
referenced alias is a node (not when all are edge aliases like r1, r2).
Fix both compilation entry point and _lower_projection_chain to use
consistent all_are_edges check.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Add 3 edge-case tests: star graph (1->N bindings), bidirectional edges
(correct src/dst swap), duplicate edges (one binding per edge).

GPU validation: all tests pass on both pandas and cudf (dgx-spark,
graphistry/test-gpu:latest container). The bindings table merge
works correctly with cudf DataFrames.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

…#982)

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Document that WITH multi-alias scalar projections are not yet
supported (goes through a different code path than RETURN).
12 total multi-alias tests now.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>