Fixes for zero-length traversals

[jeffreylovitz]

This PR fixes 2 misbehaviors:
The query:

MATCH (a)-[]-(:L{})<-[*0]-(b:L) RETURN a

Caused the server to crash.

The above query produced a different execution plan than the logically equivalent variant:

MATCH (a)-[]-(:L)<-[*0]-(b:L) RETURN a

(The second query does not have the empty property map.)

However, there are still problems evaluating 0-hop traversals. If the intermediate node does have filters, as in:

CREATE (:L {v:1})-[:E]->(:L {v: 2})<-[:E]-(:L {v: 3})
MATCH (a)-[]-(:L{v: 2})<-[*0]-(b:L) RETURN a

No results will be returned.

The difference can be seen in the execution plans:

redis-cli GRAPH.EXPLAIN G "MATCH (a)-[]-(:L)<-[*0]-(b:L) RETURN a"
1) "Results"
2) "    Project"
3) "        Conditional Traverse | (b:L)->(a)"
4) "            Node By Label Scan | (b:L)"

redis-cli GRAPH.EXPLAIN G "MATCH (a)-[]-(:L{v: 2})<-[*0]-(b:L) RETURN a"
1) "Results"
2) "    Project"
3) "        Conditional Traverse | (anon_1:L)->(a)"
4) "            Conditional Traverse | (b:L)->(b:L)"
5) "                Filter"
6) "                    Node By Label Scan | (anon_1:L)"

[swilly22]

Great addition, zero hop is confusing and easily can lead into issues, one question request.

[swilly22]

Nice!

[swilly22]

Can you explain why this loop condition was changed?
I would really like to keep the original version as it is straightforward and easy to reason about