fix(reconcile): retry retryable Neo4j errors when creating constraints

[rahlk]

Closes #11.

Release run 34004180477 failed at the gate with create constraint: Neo.TransientError.Transaction.DeadlockDetected: BoltStore.createConstraints used autocommit session.Run and returned the first error, so a retryable transient error from two concurrent writers (in CI, internal/reconcile and tests in parallel against one service; in production, two caniac --emit neo4j processes) aborted the generation.

Each CREATE CONSTRAINT ... IF NOT EXISTS now runs inside session.ExecuteWrite, so the driver retries retryable errors with its bounded jittered backoff (MaxTransactionRetryTime, 30 s default). Non-retryable errors are still returned wrapped before any data write; the session close-error handling is unchanged. Neo4j 5 accepts schema commands in a managed transaction (probed live).

Tests: three unit cases through a fake driver/session against the real createConstraints (retry after deadlock succeeds with zero autocommit calls; non-retryable error returned once, no retry; cancelled context stops). RED before the fix on the first and third. Five consecutive parallel go test ./internal/reconcile ./tests runs and one go test ./... green against a live Neo4j 5; gofmt, go vet, git diff --check clean.

Known trade-off: an unreachable server during constraint creation now waits up to the driver's retry window before failing.