Retry on certain query failures for managed transactions (#190)

When using `Graph::run`, `Graph::run_on`, `Graph::execute`, or `Graph::execute_on`, the query might be retries for a while.
The logic is that all errors with the `Transient` error class as well as a few other error classes are considered retryable.
This catches errors during a leader election or when the transaction resources on the server (memory, handles, ...) are exhausted.
Retries happen with an exponential backoff until a retry delay exceeds 60s, at which point the query fails.

Transaction created with `Graph::start_txn` or `Graph::start_txn` are *not* retried automatically, only managed transactions are.

Cherry-Picked-From: #186

Author: knutwalker

.github/workflows/checks.yml

@@ -26,7 +26,6 @@ env:
   MSRV: 1.75.0
   HACK: hack --package neo4rs --each-feature --exclude-features unstable-serde-packstream-format,unstable-bolt-protocol-impl-v2,unstable-streaming-summary
 
-
 jobs:
   check:
     name: Compile on MSRV