RocksDB memory configuration is based on static and not the actual number of running partitions processors

[tillrohrmann]

Currently, we base our RocksDB memory budget on a statically configured number of partitions (num_partitions_to_share_memory_budget in StorageOptions). With the explicit provision step, users might provision the cluster with a different number of partitions than what was configured. Moreover, if we don't run all partitions on every worker node, then this number might even change again. Hence, we might configure the memory budget assigned to a column family currently incorrectly. The bad things could be OOMs (if we set the value too high) or not using the available memory as efficient as possible.

[tillrohrmann]

fyi @AhmedSoliman