Add Bandit to tests in /integration

[mtrudel]

No description provided.

[mtrudel]

These tests run quite a bit slower than previously because we're starting / stopping the underlying server via a setup block instead of a setup_all (ie: we're doing it for each test). Cowboy seems to have a stock 5 second waiting period at the end of every test as a result (previously, it only waited once, at the end of the entire file).

We don't really need to be doing this this way, but fixing it properly will be a bit more surgery than what's here. I'll get it fixed up.

[mtrudel]

Finally had a chance to look at the Cowboy slowness. Didn't / couldn't solve the underlying issue with Cowboy taking 5 seconds to terminate at the end of each test set. Moreover, we need to keep the Cowboy startup in a setup_all call (otherwise it gets started/stopped on every test, which adds 5s to every test instead of just to the entire test suite).

As a result, the structure of these tests is a bit oddball, with the tests themselves macro'd into discrete Bandit/Cowboy test modules. The vast majority of the noise in this PR is just indenting to accommodate the macro-ing of the tests.

Would love a better way to do this; if anyone has ideas I'm all ears.

Also, there's one failing setup in test/phoenix/integration/websocket_channels_test.exs that I can't figure out how to resolve. Hopefully more eyeballs make it transparent.