update tarballs in support to load_sample

[neutrinoceros]

fix #101
Most of the fix is to repackage tarballs, so here are the new ones:

dataset	link
Sedov3D	http://use.yt/upload/db6a6b28
ExodusII_tests	http://use.yt/upload/ca1842b0
GaussianCloud	http://use.yt/upload/c3e5eb68
sedov_tst_0004	http://use.yt/upload/d7729706
xray_fits	http://use.yt/upload/808c3513
TurbBoxLowRes	http://use.yt/upload/d4393bcf
ParticleCavity	http://use.yt/upload/dbc6a778

note that in most cases I changed the extraction directory to match the tarball name (in some cases I added a directory layer), but in one case I instead renamed the dataset to match the existing extraction directory (A2052 -> xray_fits), because I find that we might as well use the more descriptive name as a key since it already exists.
I'll open a companion PR on the main repo's side to reflect this change and update the hashes. (update: this is yt-project/yt#3334)

This PR exhausts the list of improperly formed tarballs, it will not be necessary to open new ones in the future

[Xarthisius]

N00b question: why aren't we fixing load_sample to support more cases instead?

[neutrinoceros]

Well my reasoning is that load_sample is very hard to properly test in CI and it's already filled with magic (i.e. special casing that feel pretty brittle). What I'm trying to accomplish is to have the function do one thing and do it well, instead of breaking apart in every way possible.
Again, I'm so sorry that this is taking so much of your time (as well as mine) :(