The new memory tracking algorithm in HashAgg is probablistic. So we found in spill_to_disk.js we aren't throwing errors in the similar cases when allowDiskUse = false. If we set `internalQuerySlotBasedExecutionHashAggMemoryUseSampleRate` to one we have parity with reporting memory limit errors with the classic engine, but this didn't fit well into the testing framework (we would have to move spill_to_disk.js to no passthrough) and loose coverage for the sharded passthroughs.

This fix will check the memory usage in HashAgg for every 100 insertions of the hash table or the random coin flip.