Cache a fixed amount of sequence numbers in the PartitionStore for deduplication

[tillrohrmann]

With #147, the PartitionProcessor reads for every incoming shuffle message the current sequence number for the producing partition id. This will entail for every such message a RocksDB read. In order to avoid too many RocksDB reads, we could cache a fixed number of sequence numbers via a LRU cache. That way we could potentially save a lot of RocksDB reads.