Make dataset loading more consistent

These changes were motivated by
[#1283](#1283) which arose
as we used different code paths for loading a dataset viz and a
narrative.

Here we represent each dataset by a `Dataset` object. This is used for
stand alone datasets, each dataset in a tangletree, and each dataset in
a narrative. Each dataset instance describes the various API endpoints
of datafiles for each dataset, manages fetching of these datafiles and,
where appropriate, can dispatch data to update redux state.

This has been tested on various single datasets, tangle-trees, and
narratives in this repo. Notably, this commit breaks narratives with
multiple datasets; this will be fixed in a subsequent commit to
reflect Eli's work in PR #1312.

narratives/test_sidecar-files.md

@@ -44,6 +44,7 @@ A tanglegram of Vic and H3N2 makes little sense, but we should see the trees lin
 
 The left tree (Vic) should be coloured teal. When reloading the page this should initially flash grey.
 
-We should also see _both_ the entropy and the frequencies panel. The latter may take a moment to arrive when refreshing the page.
+The URL specifies that we should also see _both_ the entropy and the frequencies panel,
+however this functionality does not yet exist in Nextstrain narratives when viewing a tangletree.
 
 #### Checklist: refresh the page from this slide