Improved Sampling Support

[domoritz]

For large datasets, sampling is a valuable strategy for creating smaller datasets that can be visualized and filtered much more rapidly. Mosaic supports sampling indirectly, as one can create a table of sampled rows and visualize that. Better support might provide explicit constructs for managing samples, so that analysts can interactively move between sample levels while preserving parameters/selections. One avenue may be to add a Coordinator-level abstraction for sampled tables. Careful attention is necessary to make sure any new abstractions play well with existing caching, consolidation, and data cube indexing methods; for example, different sample levels should probably correspond to different underlying table or view names.

[domoritz]

When we have a sample, we could also stream progressively improving results to the visualization clients.