Unify Fermi-LAT datasets used in Gammapy

[adonath]

This PR addresses #1827. It replaces all the fermi datasets currently used in the Gammapy tests, docs and tutorials with a single 3FHL example of the Galactic center.

There is one exception: the fermi_lat.ipynb tutorial, which shows how to prepare a dataset from Fermi-LAT allsky data. Replacing the data there with the prepared dataset does not make sense, as the tutorial would only show loading and fitting the data.

[Bultako]

@adonath
If I understand correctly, once this PR merged the only fermi-lat datasets needed in Gammapy would be the new ones that are added here plus those present in https://github.com/gammapy/gammapy/blob/master/tutorials/fermi_lat.ipynb?

We talked recently about downloading versioned datasets with gammapy download. I have added this functionality in PR #1988 commit c4a3456

[Bultako]

@adonath
In the file https://github.com/gammapy/gammapy/blob/master/tutorials/notebooks.yaml we can declare the datasets needed for each notebook, so gammapy download can fetch the needed datasets to reproduce all the notebooks or a specific single one.

I think this file needs to be modified to incorporate the changes made in this PR.

[adonath]

@Bultako Thanks, you're right! I'll add the change now...

[Bultako]

@adonath
So, are you ok to remove DatasetFermi2FHL, DatasetFermiSurvey and DatasetFermiLat from https://github.com/gammapy/gammapy-webpage/blob/gh-pages/make.py#L194-L206? Then we could generate a smaller JSON for the datasets https://github.com/gammapy/gammapy-webpage/blob/gh-pages/download/data/gammapy-data-index.json that will eventually become a gammapy-0.10-data-index.json when releasing.

[adonath]

@Bultako Yes, I'm fine with removing those. I can prepare a PR for this later...

[adonath]

The remaining Travis fail is unrelated. I"ll merge this PR now.