[19.03 backport] Dynamically register kubernetes context store endpoint type.

[silvin-lubecki]

Backport of #1890 for 19.03

Further reduce amount of transitive vendoring required by vendorers and builds on (includes) #1887. With this my private project no longer vendors anything from k8s.io. diffstat (in addition to #1887, not cumulative) is:

 407 files changed, 87 insertions(+), 76812 deletions(-)

And it dropped transitive vendorings of

github.com/docker/{compose-on-kubernetes,spdystream}
github.com/elazarl/goproxy
github.com/evanphx/json-patch
github.com/gogo/protobuf
github.com/golang/groupcache
github.com/googleapis/gnostic
github.com/google/{gofuzz,uuid}
github.com/imdario/mergo
github.com/json-iterator/go
github.com/modern-go/{concurrent,reflect2}
github.com/mxk/go-flowrate
github.com/onsi/gomega
github.com/spf13/pflag
golang.org/x/net
golang.org/x/oauth2
gopkg.in/inf.v0
k8s.io/{api,apimachinery,client-go,klog,kube-openapi,utils}
sigs.k8s.io/yaml

for a ~2M reduction in my vendor dir (which is still 27M, but is now dominated by golang.org/x/* (mostly sys, but text too), github.com/containerd/containerd, github.com/docker/docker).

There's a few things I'm unsure of:

• The mechanism of EndpointDefaultResolver interface being implemented by the concrete endpointmeta types feels a bit iffy. Might be better to expand NamedTypeGetter to have an additional method alongside the typeGetter.
• The special casing (and necessary open coding) of kube logic in ResolveDefaultContext is rather gross (a layering violation of sorts). I justify it (sort of) in that while the kubernetes endpoint is notionally optional there is a coupling via the stackOrchestrator options which means the core needs to be somewhat aware. At first I had it where it would pass the stackOrchestrator as an argument to kubernetes.ResolveDefault, I wasn't terribly happy with that either but maybe it is better?
• I really wanted to pass a Cli to ResolveDefaultContext not a store.Config and use cli.contextStoreConfig (and cli.configFile) but NewAPIClientFromFlags doesn't have a cli to use.
• I don't much like exporting DefaultContextStoreConfig() and ResolveDefaultContext() just for use in unit tests but I couldn't see another way. Similarly/Relatedly I'd have preferred not to need to split TestDefaultContextInitializer into a kube part, but couldn't avoid it while also avoiding the circular imports.

[GordonTheTurtle]

Please sign your commits following these rules:
https://github.com/moby/moby/blob/master/CONTRIBUTING.md#sign-your-work
The easiest way to do this is to amend the last commit:

$ git clone -b "19.03_backport-reduce-vendoring-impact2" git@github.com:silvin-lubecki/cli.git somewhere
$ cd somewhere
$ git rebase -i HEAD~842359068512
editor opens
change each 'pick' to 'edit'
save the file and quit
$ git commit --amend -s --no-edit
$ git rebase --continue # and repeat the amend for each commit
$ git push -f

Amending updates the existing PR. You DO NOT need to open a new one.

[codecov-io]

Codecov Report

Merging #1907 into 19.03 will decrease coverage by 0.04%.
The diff coverage is 44%.

@@            Coverage Diff             @@
##            19.03    #1907      +/-   ##
==========================================
- Coverage   56.72%   56.67%   -0.05%     
==========================================
  Files         309      309              
  Lines       21678    21707      +29     
==========================================
+ Hits        12297    12303       +6     
- Misses       8483     8505      +22     
- Partials      898      899       +1

[thaJeztah]

@silvin-lubecki can you try force-pushing again? I think the dco-signed check got confused because it was originally opened against the wrong branch 🤔

[thaJeztah]

LGTM

[albers]

As the PR does not contain changes to bash completion any more, I removed myself from the reviewers.

[andrewhsu]

LGTM