Debug tracing for ANN search

[changhiskhan]

oftentimes we struggle to explain the results combine from the ANN index. To improve the developer experience, what if Lance's nearest neighbor API supported an option to return debug information about how the result was achieved?

e.g., if you call dataset.to_table(..., nearest={..., debug=True}), then you'll get additional columns back in the results like:

1. the ivf partition id this vector came from
2. the raw score (if refine_factor was not none)
3. the list of pq sub vector id's

With this information, combined with the index information itself from #617 we can make it very easy to understand and debug where results come from.