0.3.0-beta3

Now supports chunking of very large nodes. Splits them into 10,000 char chunks. Very rough and needs to be improved with proper language-aware sentence splitting. But, this prevent large nodes (like 2hr long meeting transcripts) from breaking the embedding system.