Sketchnu

This gives more control to the user when using helpers.parallel_add()

The helpers.parallel_add() was updated to be able to leverage the changes in v1.2.2. The generator function, process_q_item, now needs to yield a tuple. First element is either a Dict[bytes, int] representing the keys and their associated number of times to add them to the sketch or just an Iterable[bytes]. **NOTE** Using Dict is likely faster, but is data dependent. The second element is an int that is the number of records processed. This is so the count-min and heavy hitters sketches can record the number of records processed, as they did before.

Allow adding the same key multiple times to the different sketches. Previously this was only done by repeatedly calling add(key). But now you can do add(key, value=50). This can be a significant speed up if the same key needs to go in multiple times. This capability allows the sketch.update() to now also take a dictionary (or collections.Counter) {key1: 50, key2: 25, key: 10}.

Utilizes numba's parallel=True & prange to speed up the merging of count-min sketches and heavy-hitter sketches. This parallelizes the ``for`` loop through the number of rows in the underlying 2-d arrays (i.e., the ``depth`` which defaults to 8).

Changed the parameterization of the HeavyHitters to better match with the theoretical guarantees provided by the paper. Also greatly expanding the testing to include testing for the theoretical guarantees and now use pytest to conduct the testing. Lastly reorganized the test scripts into the tests/ directory to better follow conventional project structures.

Initial release