Drop support for python < 3.6

[mgaitan]

I've been working in a couple of refactors and I feel it's time to stop supporting both python 2.7 and python 3.5, and ensure support to python 3.7 (see #1018), 3.8 and any new python release in the future.

As you may know, python2.7 reaches its end-of-life date starting 2020, and 3.5's one is the next september. Several of the most relevant libraries of the python ecosystem (e.g, django and numpy, just to mention few) are already python3 only since a while.

This would simplify the code and the tests a bit (not a big deal, to be sincere), but most importantly, it'll allow us to start using new idioms/features of the language without hacks, conditional requirements, etc.

[Zulko]

I am in favor. Anyone against?

[tburrows13]

According to pypinfo, 10% of downloads in the last 30 days from PyPI are on Python 2. On https://pypistats.org/packages/moviepy it hovers at around 20% (same source data as pypinfo, so probably just because of those massive spikes).

I agree that it is worth doing. Python 3.5 I'm not so sure if the tiny benefits outweigh the decreased support. I don't really mind though.

For completeness, here are some PyPI stats from the last 30 days:

version	percent	download_count
1.0.1	73.94%	146,036
0.2.3.2	13.82%	27,296
1.0.0	5.31%	10,481
0.2.2.11	4.83%	9,538
0.2.3.5	1.70%	3,359
0.2.2.13	0.20%	393
0.2.3.4	0.07%	132
0.2.3.1	0.06%	125
0.2.1.6.3	0.04%	75
0.2.3.3	0.04%	71

python_version	percent	download_count
3.6	63.88%	127,917
3.7	11.74%	23,507
2.7	10.84%	21,699
None	6.01%	12,041
3.5	3.95%	7,906
3.8	3.50%	7,002
3.4	0.07%	148
3.9	0.01%	16
2.6	0.00%	2
3.2	0.00%	1

system_name	percent	download_count
Linux	84.28%	168,762
None	8.92%	17,857
Windows	5.00%	10,011
Darwin	1.80%	3,598
FreeBSD	0.00%	2
OpenBSD	0.00%	2
MSYS_NT-10.0-18363	0.00%	1
CYGWIN_NT-10.0-18362	0.00%	1
MSYS_NT-10.0	0.00%	1
CYGWIN_NT-10.0-WOW	0.00%	1

[keikoro]

Sounds good to me.

[tburrows13]

With the merging of #1106, this PR can now be closed.