Drop writes before the retention policy time window

[jwilder]

#6635 seems to be caused by writes into the past that triggers new shard groups and shards to get created. These shards would end up getting removed when retention policy enforcement runs, but they can create a lot of overhead and memory pressure before retention policy enforcement runs. If the shard duration is somewhat small (1d or less), many shards can get created in this time.

We should drop writes if they would end up creating shard groups that are already expired as defined by the retention policy.