[graph_memory]: improve delete/add graph memory

[spike-spiegel-21]

Description

Flow of graph memory addtion:

1. Extracts all the entities/nodes and their types mentioned in the query. (Returns: dictionary of {nodes:type})
2. Eshtablish relations among the extracted nodes. (Input: list of nodes(that does not contains types). Returns: list of source -- relationship -- destination)

(Note: The above 2 LLM calls are made to ensure that all the facts are retrieved from the query/data. The single LLM seems to be overloaded if we expect source (and their types) -- relationship -- destination (and their types))

3. Search similar nodes and their respective inbound and outbound relations.
4. LLM call to find what relationship we want to delete based on the query/data. (it returns the relations that we have to delete

Now we have a list of entities (source -- relationship -- destination) that we need to add and delete.

5. Delete relationships
6. Add relationships

• Find whichever node is common (source and destination) via similarity search (0.9 threshold) and then then merge it.