PySpark Cleanup of _setDefault with seed=None

Several places in PySpark ML have Params._setDefault with a seed param equal to None. This is unnecessary as it will translate to a 0 even though the param has a fixed value based by on the hashed classname by default. Currently, the ALS doc test output depends on this happening and would be more clear and stable if it was explicitly set to 0. These should be cleaned up for stability and consistency.