Implement Relevance Scoring in Episodic Memory

Nobody has claimed this yet.

**What's the problem this feature will solve?**
The current EpisodicMemory module successfully implements recency and LLM-graded importance scoring, but is missing the relevance logic that is needed, since the entire memory architecture is inspired from the "Generative Agents" paper, We need to implement vector similarity scoring to complete the "Generative Agents" retrieval architecture.

**Describe the solution you'd like**
Adding a relevancy based scoring logic to the entire file based on a query string which is obtained from the observations stored.

**Are you planning to open a PR for this?**
Yes I'm working on a PR for this issue :)

Start in the EpisodicMemory module and read the existing recency and LLM-graded importance scoring logic. Trace how observations produce the query string, then inspect the retrieval flow to determine where vector similarity belongs. Done means relevance scoring is integrated with the existing retrieval architecture described in the issue.

Written by the indexing model from the issue text.