Simplify pytest remote data mock setup

[cdeil]

This is an attempt to simplify the pytest remote data mock setup that was introduced in #98.

Please see #98 (comment) and #98 (comment) as well as the
docstring of _mock_remote_data in this PR where I try to explain how it works and why it's an improvement.

[cdeil]

@eteq Can you please review this?

[eteq]

@cdeil - I'm not sure this actually solves the problem #98 was initially meant to fix. With this version, you always turn on mocking of --remote-data is off, and not if it's on. That means there's no way to test the mocked versions if remote_data is on.

My thinking is that that's a feature, not a bug. That is, it's better to write remote_data and non-remote_data tests as distinct and different entities. Then it's clearer what the origin of failures are: the actual remote_data behavior (i.e. downloads), or the other pieces that just use remote downloads incidentally.

To be more concrete, here we should have tests that actually test if from_name downloads the correct result, in addition to tests (mostly doctests) that use from_name just to make things easier. We want the latter to use the mock results regardless of remote_data status, and the former should never use mock data. The original implementation does that, but in this version, if you turn remote-data on and SIMBAD is down, from_name will yield tons of spurious failures, rather than just one. So I would advocate keeping pytest_runtest_setup as it was in #98.

The rest of it are good changes, though.

[eteq]

After a lengthy discussion on slack, @bmorris3 broke the tie and said we should switch to this behavior. So I'll merge this PR, and after this, --remote-data means from_name always goes to the internet, regardless of whether a test is marked remote_data or not.