Python coding cleanups

[jllanfranchi]

• Run the code through pylint to get general feedback on Python coding conventions and also basic bugs you might not otherwise catch
• E.g.: one example I see is you shouldn't use type(x) == T; use instead isinstance(x, T).
  Obviously this is secondary to getting the code working and performant, but it's always easier to start early rather than build up a large codebase and then check with pylint.
• FTYPE should be a single global value, and should not be changed. Function / parameter names should be all lower-case (possibly with underscores) and NOT be the same as the "globals" e.g. FTYPE. FTYPE so a function can take an argument ftype but be assigned the global FTYPE as a default, but within the function, it should make use of ftype. I.e.: func(x, y, ftype=FTYPE) is workable. func(x, y, FTYPE) or using FTYPE within the function if FTYPE is ever to be changed is not workable.
• n_events argument indicates "number of events" yet there is also a number_of_events argument (GPUHist class / method(s))
• Think through the logic of testing and implement it modularly; as it stands, there's a lot of copy-paste code in there, and things doing odd things with peculiar arguments to functions (e.g. creating an array NOT on the gpu (create_array(device_array=False)) returns a second value (usually reserved for a gpu-based array), but it's just a regular numpy array... then your subsequent code assignes that value to a variable d_input_data and might or might not work with it).

[mhieronymus]

• Using pylint
• use isinstance (already changed it on many occasions)
• Don't mess up global and local variables (FTYPE)
• Rename n_events and number_of_events (n_events is now sample and number_of_events is now n_events because it makes more sense than before)
• Rework the logic of the code.