Add WARN for large resultsets

Similarly to how we warn on excessive tombstones or large batches - it would be useful to warn if we come across a large row over a specific threshold.

If not too intrusive it could provide us with a smoking gun for gc thrashing type issues.

To put some meat on the bones - I've recently seen someone store a 1Gb blob and then try to work round it by splitting it into 3 columns of 330 Mb each. Neither gave particularly good performance.

On the opposite side of the coin - a large number of columns grabbed at once can also lead to problems.