fix: prevent shared driver state mutation in multi-group FalkorDB usage

[StephenBadger]

Summary

• add_episode() and add_episode_bulk() mutate self.driver and self.clients.driver when switching FalkorDB graphs for different group_ids. In a multi-tenant server with one shared Graphiti client, writes to one group silently redirect all subsequent reads for other groups to the wrong FalkorDB graph database, returning empty results.
• Root cause: FalkorDB uses a separate graph database per group_id. add_episode() clones the driver to point at the target group's graph but stores the clone back into self.driver/self.clients.driver, permanently changing which graph the shared instance targets. Concurrent or interleaved operations targeting different groups corrupt each other's queries.
• @handle_multiple_group_ids decorator only cloned drivers for len(group_ids) > 1, so single-group searches used whatever stale driver state the last add_episode left behind.

Fix

• add_episode() and add_episode_bulk() now create local driver/clients via clone() and model_copy() instead of mutating shared instance state
• All private methods (_extract_and_resolve_edges, _process_episode_data, _get_or_create_saga, _extract_and_dedupe_nodes_bulk, _resolve_nodes_and_edges_bulk) accept optional driver/clients params so the local clone flows through without touching self
• @handle_multiple_group_ids decorator now clones the driver for single group_id queries too (>= 1 instead of > 1)

Reproduction

1. Create a Graphiti instance with a FalkorDB driver
2. Call add_episode(group_id="group-a", ...) — driver now points to group-a graph
3. Call add_episode(group_id="group-b", ...) — driver flips to group-b graph
4. Call search_(group_ids=["group-a"], ...) — returns empty results because driver is querying group-b graph

[danielchalef]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[StephenBadger]

I have read the CLA Document and I hereby sign the CLA

[StephenBadger]

recheck