AnalysisNode - cache counts + grid by input + hash of config

Currently, we increment node versions, which means if you flip back and forth between config for a node, you have to calculate it each time.

If instead we stored this against the parent input and node config (or even just the hash of the SQL - which in effect is the same thing) - then we could switch back and forth between config quickly.

This would also mean that shared data would be cached, for instance if we decided to make a demo on the front page, it would be almost instant after it had been used enough.

Another option is to make a caching layer between a grid and the query, so we could eg query 5 pages worth of data, then pull out the data manually (1st test whether this is a big performance gain - ie time to page through 5 pages vs 1 page 5x as big)