Currently our logic for retryable timeseries updates assumes we get a response from a shard and doesn't know how to handle local errors: https://github.com/mongodb/mongo/blob/cf78d87ee503ecb877d261c9138d786b0b3b399f/src/mongo/s/write_ops/bulk_write_exec.cpp#L384-L385

The logical change should be simple: call through to the new 
BulkWriteOp::processLocalChildBatchError method added in SERVER-80481 in case of a local error. 
But we should add tests along with that to confirm it works as intended.