remove support for Python 2.x, require python 3.7 or later

[ctb]

This is the issue that @luizirber has been waiting for for years!!

TODO --

Make necessary documentation (install docs, requirements docs) changes.
Adjust conda recipe
Remove python 2.7 builder on CI.
Remove all of the future statements.
change/fix Enum import stuff

remove compat code, including

sourmash/_compat.py:PY2 = sys.version_info[0] == 2
sourmash/lca/command_index.py:    if sys.version_info < (3, ):
sourmash/signature.py:                if sys.version_info >= (3,):

[ctb]

also #1043

[luizirber]

I think we should move the minimum supported Python 3 according to the NumPy policy:
https://numpy.org/neps/nep-0029-deprecation_policy.html
Although... this would mean dropping 3.6 yesterday =P

OK, maybe making it 3.6+? I don't see why support 3.4, maaaybe support 3.5, but bioconda only builds for 3.6 and 3.7 (and slowly transitioning into 3.8).

[ctb]

I'm good with following that NEP. Any downsides?

(The availability of conda install really makes this a no-brainer for me.)

[luizirber]

I'm good with following that NEP. Any downsides?

It's a faster deprecation cadence than what we had previously, but other than that I really like the reasoning in the NEP.

One point is that we don't really support anything besides the latest released sourmash version, which is different than what NumPy does.

(The availability of conda install really makes this a no-brainer for me.)

We might also consider what are the default Python versions on MacOS and Linux (Ubuntu LTS?), because that might make it harder to install in non-conda places. Although at this point that's "advanced usage", and it's up to the user to install the correct Python?

[ctb]

Yes, I think it is ok to have a reduced/minimal support policy. At this point I'm still not sure how many people are using sourmash, and in the absence of that and more dedicated funding I don't want to complicate our lives for this unknown! I also don't think software support is a differentiator for adoption here, to be blunt...

Also, I would be totally fine with releasing maintenance versions of 3.x if we discover major bugs...