Optimization while indexing, and sending context.

[vrajhm]

1. Reduced the context to the top 8 snippets with semantic search. This can be changed if you prefer bigger context.
2. Fixed an typo on the Jinja error.
3. Chunking the embedding, allowing larger repos to be embedded for less memory. Drawback: it is slightly more time consuming for smaller repos.