🚀 The feature, motivation and pitch

Right now ColBERT always uses PLAID for indexing documents, which is not optimal when the number of documents is small.
In chat-rag, the ColBERT retriever already allows PLAID and encoding memory so the task would be to integrate this retriever into the back ColBERT actor, so we have better indexing times when using small knowledge bases.

Additional context

No response