Made it possible to skip chunking

[whitead]

Made a few changes to get closer to using Gemini, where we do not chunk and work with whole documents.

• a chunk_size of 0 will trigger skipping chunking
• deferred embedding documents until retrieval, to avoid embedding docs all the time in tests (or in cases like when you're not doing vector retrieval)
• added a small helper function to truncate before embedding, in case a text is too long