Add additional correlation tests to covidcast

[capnrefsmmat]

#217 adds a starter set, but we need more. These can be added piecemeal over many PRs, so they're good starter tasks for anyone who'd like to try working on the R package.

• Bubble plots. Particularly the behavior when counties have missing data, have data present but have 0 counts, or have nonzero data. They should be tested for both county and state levels, and also when there are megacounties.
• More on correlations. We do not currently test by = "time_value" at all, nor do we test different correlation types (Spearman, etc.). And when Handle warnings raised in covidcast_cor #33 is fixed, we should test it.
• aggregate_signals takes several different formats for dt depending on how many data frames are provided; we should test all of these formats.
• covidcast_signal should be tested (using mocks) to ensure it formats its output correctly.
• covidcast_signal's warnings/error messages (when a signal or day is missing) should be tested so they don't regress (see e.g. msg has been renamed to message`  #251)
• It should be possible to call covidcast_signal on signals with no metadata, provided you provide start_day and end_day.

ryantibs you're welcome to add other test ideas to this list for others to take on.

[ryantibs]

Looks good to me, thank you Alex. Atm I can't think of much else to add. But will leave comments here as I keep working on development to see if new things come to mind.

Nb @jacobbien @bnaras @dajmcdon on the evalcast side: you can also look at #217 to what Alex already did and how he set it up through testthat.

[capnrefsmmat]

The only thing remaining here is to add additional correlation tests, so I'm renaming this.