Drop py3.5 support, support py3.8, fix lint test

[carver]

What was wrong?

fix #96

Also, drop py3.5 support (breaking change, requires major version bump, of course!)

How was it fixed?

Update tox, circle-ci etc.

While doing that, notice that the lint tests were probably broken. Hm, nope, the lint tests were working befdore! It's a mystery to me why, I really thought that tox's testenv didn't support partial matches. In fact, it came up pretty recently in another project and the lint test was getting skipped, which I thought was because of a partial-match issue...

Also, had to upgrade hypothesis to work with py3.8. This meant that warnings about long tests turned into errors, so some tests had to be made shorter and/or the deadline extended.

Cute Animal Picture

[pipermerriam]

Any reason you chose not to run the linter across the different python versions? Otherwise 👍

[carver]

Any reason you chose not to run the linter across the different python versions? Otherwise

Well at first I thought that lint wasn't running at all because of the [testenv:lint] definition, which does not trigger in the environment py36-lint for example. This was a problem in ethereum/eth-keys#72 which definitely was not running lint tests before. So I was surprised to find out that py-trie was running lint tests despite this partial match problem.

After some digging, I found: .circleci was running the lint environment instead of py36-lint:

py-trie/.circleci/config.yml

Lines 47 to 58 in 3be5646

	py35-lint:
	<<: *common
	docker:
	- image: circleci/python:3.5
	environment:
	TOXENV: lint
	py36-lint:
	<<: *common
	docker:
	- image: circleci/python:3.6
	environment:
	TOXENV: lint

Which means that lots of confusing things were happening:

1. a local run of tox -e py36-lint would give you a different result than the circleCI job py36-lint (which was really just tox -e lint but in a py36 container)
2. there were a bunch of listed environments in tox that were never run in CI
3. the only thing forcing the correct python was in the circleCI config instead of the tox one

So...

I think the right answer is to just stick with the project template standard and only run one linter:
https://github.com/ethereum/ethereum-python-project-template/blob/e38d2349dd9941ac38cae5ab0402a912092fce2d/tox.ini#L4

If the linter itself is giving different results for the same version of flake8 but different versions of python, then I think we can just recognize that as a bug in flake8. That reduces our CI time & noise. I don't think we get much benefit from re-checking lint in a bunch of different pythons.