Shared relationship variable across pattern parts falls back to a cross join #58

Problem

A relationship variable reused in two comma-separated pattern parts should join those parts on that one relationship. Instead, the planner joins them with a cross join and returns every combination.

Reproduction

import pyarrow as pa
from lance_csr import CypherEngine, GraphConfig

config = (
    GraphConfig.builder()
    .node_table("Person", "id")
    .rel_table("KNOWS", "Person", "Person", "src", "dst")
    .build()
)
people = pa.table({"id": [1, 2, 3, 4], "name": ["A", "B", "C", "D"]})
engine = CypherEngine(config, {"Person": people, "KNOWS": pa.table({"src": [1, 2], "dst": [2, 3]})})
engine.execute(
    "MATCH (a:Person)-[r:KNOWS]->(b:Person), (c:Person)-[r]->(d:Person) "
    "RETURN a.name AS name ORDER BY name"
).to_pylist()

This returns 8 rows (A four times and B four times); openCypher returns 2 (A and B). The result is the same with and without use_csr=True.

Cause

The logical planner chooses a cross join for the two pattern parts before join keys are inferred, so the shared relationship variable never contributes its endpoint columns as keys. infer_join_keys itself handles relationship variables correctly; a planner test covers it since #51.

Suggested fix

When comma-separated pattern parts share a relationship variable, plan an inner join on that relationship's endpoint columns instead of a cross join.