Rewrite helm-operator to use weaveworks/flux/git

The go-git package seems to introduce a host of problems, especially
with SSH, that are obviated by using the git binary.

Given a couple of modest additions to github.com/weaveworks/flux/git,
we can use that instead. Namely: read-only repos, and repo "exports"
(that is, clones that come with only the ability to look at the
files).

It also turns out that it's possible to simplify the chartsync and
releasesync packages. Given the whole picture in the existing code,
it's possible to boild it down to two mechanically simple processes:

 - chartsync looks for the charts used by FluxHelmRelease resources,
   and checks whether they've changed since the last commit;
 - releasesync installs any FluxHelmRelease resources that
   don't have corresponding releases, and upgrades any that do and
   differ.

(In fact, just the latter would be fine since new commits would result
in release diffs; but it's more _eventual_ than having both).

Since these are now fairly compact, I've just merged them into the
chartsync package. I've removed the releasesync_tests.go here, since
it tested internal machinery that no longer exists, but I ought to
replace the test coverage forthwith.

Gopkg.toml

@@ -42,10 +42,6 @@ required = ["k8s.io/code-generator/cmd/client-gen"]
   name = "k8s.io/helm"
   version = "v2.8.1"
 
-[[constraint]]
-  name = "github.com/google/go-cmp"
-  version = "0.2.0"
-
 [[constraint]]
 	name = "github.com/justinbarrick/go-k8s-portforward"
 	version = "v1.0.0"