kvserver: allocator should use io threshold instead of l0 sublevels #85084
Description
io threshold encapsulates the io overload a store is experiencing. We currently use the smoohted (max over 10 minutes) L0 sublevel count to exclude stores from rebalance targets.
This should be swapped with the io threshold instead, as this encapsulates the files and sublevels together.
cockroach/pkg/kv/kvserver/store.go
Lines 945 to 948 in d7b901d
The value used should remain smoothed, using the swag utility on the receiver end to track the maximum.
Added in #78608
Jira issue: CRDB-18032