One important difference between Classic and SBE is how engines handle spilling to disk in $group.

Classic spills to files on disk and uses external sort.

SBE uses B-trees and a separate spilling WT instance.

We should have a benchmarks that sends $group query that requires spilling and monitors performance and spilling metrics (spills, spilledRecords and spilledBytes).

We must ensure that SBE is at least as efficient as Classic in common cases and optimizer this if needed.