If we have a write on a sharderd and unsharded collection in the same database and we have a mismatched db version, the write should fail on the unsharded write, but because bulk_write.cpp eagerly sets shard role for all involved namespaces, it will fail even on the sharded write, that should not care about db version.

This is more of less fine, but it misattributes the failure to the incorrect write index, causing a bug in bulk_write_exec.cpp