[SYSTEMDS-3891] Improved Stream Handling and PCA support

[janniklinde]

This patch introduces various bugfixes and improvements to stream handling to address synchronization issues leading to deadlocks in specific cases (we now reliably guarantee that the finalizer callback is invoked after the last consumer finished execution). Further, the Tee instruction was modified to be introduced by statement level rewrites to keep track of transient reads / writes which led to issues with the previous DAG rewrite.

Finally, we introduced basic block tracking to remove items from the OOCEvictionManager once they have no pending consumers. To achieve that, we track stream references and once no variable holds the underlying CachingStream we mark it as deletable (because then we know, that no additional stream consumer will be added). Blocks from deletable streams can safely be removed from the cache once their consumption count reaches the final number of registered consumers.

Note that we modified the VariableCPInstruction to access variable remove and copy events.

[mboehm7]

LGTM - sorry for the delay, and thanks for the various improvements @janniklinde. These changes are great progress towards full algorithm support. During the merge, I fixed minor issues related to unnecessary imports, missing synchronization of overwritten methods, and duplicated but commented code. For the PCA test, it would be good to add the result comparison with a future PR.