Make dataframe dependencies optional

[astrofrog]

As discussed in #428, this makes the dataframe dependencies optional. I added a CI job to make sure everything works without the optional dependencies

[m-albert]

Generally looks good to me! But there seems to be a problem in the absence of the two dependencies..

[astrofrog]

The failure was because of a change in recent versions of scipy, I've fixed compatibility with these versions now (just an issue with the tests). In short, generic_filter is pickier now and requires true scalars and doesn't accept length-1 arrays. This wasn't picked up before because scipy is pinned in the CI for other jobs.

[m-albert]

Nice, completely agree with the way you split out the dependencies.

Thanks a lot also for fixing the testing issues you found! This fixes the first 4 test issues highlighted in #427 as well. Since this is not directly related to this PR, I'll open a new PR to add a test test-latest-all-extras to cover the case of latest environment + find_objects.

Before merging, could you add a test here which makes sure the ImportError is correctly raised when dataframe is not available?