Move the direct dependency on libgit2 to an extension

[ChrisRackauckas]

This is a breaking change, and thus updated to v1.0 with a note in the README on updating.

[fredrikekre]

Since package installs are not repos anymore I wonder how useful this functionality is. It basically never works, right? I guess it works when you build your own docs in CI though.

[ChrisRackauckas]

Yes, so it generally doesn't work except for CI building of docs, which is another reason to not require it downstream.

Also, the failures in CI here are also failures on master, so there's other unrelated maintenance issues in this package which hopefully won't block this change.

[nhz2]

Please test that this change doesn't break packages' documentation. For example: https://ap6yc.github.io/AdaptiveResonance.jl/stable/man/full-index/#AdaptiveResonance.classify-Tuple{ARTModule,%20AbstractVector{T}%20where%20T%3C:Real}

And the many other uses of the METHODLIST https://juliahub.com/ui/Search?q=$(METHODLIST)&type=code

[ChrisRackauckas]

How is that related? I only touched URL. MethodList is completely untouched and still passes its tests.

method lists                 |    3            3  1.3s

[nhz2]

The url function is explicitly not part of the public API of this package so it can be removed without making a breaking change if there are no downstream problems.

[nathanaelbosch]

According to SemVer pre-1.0 releases are allowed to be breaking, so there is no actual need to make a 1.0 release I think? Any compat entry for 0.9 would need to be manually updated to 0.10 anyways to opt in to this change. And the downside of the 1.0 release is that afterwards, changes like this actually need a new breaking release to a major version.

[KristofferC]

Extensions ideally shouldn't be used like this because your code might all of a sudden break if a transitive dependency drop their libgit2 dependency.

[mortenpi]

Let's close this -- I think #167 is the simpler approach here.