feat: Neo4j slow query logger

[aintnostressin]

Adds slow query logging for Neo4j.

GraphExec trait becomes the main application interface for interacting with the Neo4j database. TracedGraph is a decorator over GraphExec that logs queries exceeding a configurable time threshold. Graph is a plain GraphExec implementation that delegates directly to neo4rs::Graph - used when slow query logging is disabled (e.g. migrations, CLI commands).

All graph queries now use a custom Query type (replacing neo4rs::Query) that carries a label for identification in logs.

Note on timing: Measured query time is wall-clock time from the application's perspective. It includes the Neo4j roundtrip, driver-level connection pool acquisition, retry logic, row streaming, and async executor scheduling delays - not just
server-side execution time.

Configuration:

• slow_query_threshold_ms - queries exceeding this threshold are logged with their label and timing breakdown (execute vs fetch).
• log_slow_query_cypher - additionally logs the fully interpolated cypher for slow queries. Has per-query overhead so should not be enabled in production by default.
• slow_query_logging - set to false to disable tracing entirely (used by migration and mock configs).

[ok300]

A single last comment, otherwise LGTM.