Add a stub test for connections the server closes while idle
Cross-driver cover for the Aura roll case: the server drops an idle pooled connection without telling the driver, and the next query should transparently get a fresh one instead of failing on a corpse.
Gated behind a new Optimization:DeadConnectionDetection feature, so drivers that have not implemented this skip it. Only the .NET driver declares it so far, in neo4j/neo4j-dotnet-driver#968.
The connection has to die while it is idle in the pool, so the script deliberately never runs to completion and the test stops the server to close it. That ordering comes from the test rather than from the wire: by the time session.close() has returned, the driver has released the connection. Nothing driver-specific is waited on and there is no sleep. A pool of one forces the second query onto the same connection rather than leaving it to chance.
Verified against the .NET driver: passing as written, and failing with ServiceUnavailableError: Unexpected end of stream when the check is stubbed out. With no feature branching in the test, that result applies to any driver rather than only ones that reset on release.