DOC/PY3: Python 3 docs example compat for range, StringIO and datetime

[phaebz]

related is #6212

Fixes examples in documentation to be compatible with Python 3. Also tested it locally under Python 2.

• Turned occurrences of (x)range() to range() for iteration and to list(range()) for list concat.
• Fixed leading zero argument(s) as in e.g. datetime(2013, 01, 27).
• Removed explicit StringIO and cStringIO imports in examples (uses suppressed pandas.compat imports) and added a note on py2/3 library reorganization.
• Fixed StringIO for binary data to the more explicit BytesIO.
• Fixed NameError for missing pd. in pd.Series.
• Added note for failing numpy.unique example, see also DOC/PY3: np.unique errors on pd.Series with mixed string/float in python3 #6229.

I am not sure about the BytesIO though. Maybe good to put a note saying this is Python 3 way to do it and users on 2.x should use the old StringIO instead. In general, should we favor Python 3 doc versions in such cases and add notes for Python 2.x users? What do you think?

[jreback]

seems ok to me...@y-p ?

[jorisvandenbossche]

Is the conversion of xrange() to list(range(..)) necessary? Why not just to range()?
This works both on both python 2/3 in a list comprehension, and seems easier.

[ghost]

as long as the pandas.compat imports are limited to wrappers for known classes
it's not a problem. Looks fine.

Wherever the code does not explicitly require a sequence for random access e.g. foo[4],
the list(range(foo)) is not required as joris pointed out.

[jreback]

yep you can just use range (if u r iterating over it)

use list(range) if you actually need a list

[phaebz]

Adapted.

[jorisvandenbossche]

forgot this one I think?

[phaebz]

Thanks for pushing my nose into it... amended.

[phaebz]

If this looks ok to you now, I could go ahead and squash...

[jreback]

go ahead and squash

[phaebz]

done.

[ghost]

@phaebz , please get rid of all the `list(range())`` where not required, I think that's
everywhere.

[jorisvandenbossche]

@y-p regarding the PR, I think it is OK now. The remaining list in list(range(1,24))+[np.NAN] is needed for being able to add the nan.
But elsewhere in the docs there will be still some unneeded list(range())s, but that can maybe be left for another PR.

[ghost]

You're right, good. merging.