simplifies transaction response parse, fixes path/nodes bug

[subvertallchris]

This fixes neo4jrb/activegraph#1125 and, in the process, simplifies our handling of responses from transactions.

Neo4j Server provides different JSON formats. When returning data from the legacy Cypher endpoint, you get an arrays of data in what they call REST format, which contains everything we need: metadata (labels, type, id) and properties. When returning from the transactional endpoint, you get... I don't remember what the default is called, but it doesn't give us what we need, so we also tell it to include the REST format separately. The result is two keys, one containing the default transaction response, the other containing the more complete response.

What we were doing was complicated. We were looking for transaction responses and then keeping track of how far along we were in a loop through the default, insufficient data, and then supplementing it with data from the other, complete REST representation. This could lead to weird problems when things didn't map as expected, like when returning paths.

The solution is really simple: as soon as possible, figure out if you're dealing with a transaction. If so, just work entirely from the REST representation and ignore the weak transaction response entirely. It simplifies a lot of code and fixes the original problem.

[subvertallchris]

This makes changes to the public API of Neo4j::Server::CypherResponse but I don't see any possible situation where those methods could be used outside of the old procedure. They really could have been marked private. I think that the resultant bugfix is significant enough to warrant release in 6.1.x. What do you think? @cheerfulstoic

[cheerfulstoic]

Agreed! I'd like to see it in 6.1.x as well. The only thing that I'd change is putting in a separate note in the CHANGELOG to have a Changed as well as the Fixed so that people can be aware of the change

[subvertallchris]

Cool, all done. Once this passes, I'll do a release.