Using tracing framework from forked controller-runtime

Just a few spans at intermediate levels.

Annotate seed node via file

It can contain shell characters such as single quote and !, so put
the contents in a file before calling `kubectl annotate`.

Interestingly there is no option in `kubectl` to take the value from
a file, so we still have to `cat` it on the command-line.

Done in http://github.com/weaveworks/cluster-api-provider-existinginfra/pull/30

Plumb Context through 'plan' steps

Doesn't do anything different as yet, but this will be useful for
tracing, or for cancelling a command.

Ignore kubectl error during multimaster test

We often get "etcdserver: leader changed" responses from kubectl, which
can be ignored.
Rather than special write special code for that one error, just log and
ignore all errors on 'kubectl get nodes'.

Remove spurious 'test.yaml'

Tidy the CAPEI image name used in multimaster_test

Add 'status' subresource to CRD definitions

* Use a const instead of repeating the name.
* Keep it named `cluster-api-existinginfra-controller` instead of
renaming it to `wksctl-controller`

* fix erroneous image name and machines filename
* ensure all test calls turn on debugging.

This is essential to fit with the controller-runtime framework.

* test ubuntu as well as centos nodes in integration tests

* fix dead code linter error

* updates for nodeos selection

* update Makefile to allow random selection of node os

* extend timeout on integration-tests-container

Update the version of weave-net to 2.7.0

Cherry-pick of 575b70c6

Co-authored-by: Simon Howe <footless@gmail.com>

* make node os configurable

* remove dead code

* add ubuntu support for upload-image.sh

* remove outdated dependency from the integration-tests-container target

* fix typo in kubeadm_init resource name

* update dependencies

* mv examples/footloose/ubuntu1804 to examples/footloose/ubuntu18.04

* switch to capei controller

* Debug integration tests (#295)

* add node debugging

* whoops

* pin the controller image

* fix log typo

* see the kubeconfig in output

* fix debugging;

* display controller logs

* more debug

* more

* more

* more yet

* sheesh

* switch controller image

* fix wksctl image lookup from manifest

* switch image for debugging

* include changes to manifest

* more debug

* switch to weavek8sops and allow for the fact that the controller is always running in debug mode

* More namespace fixes

* lint fixes
Co-authored-by: Jerry Jackson <jerry@weave.works>

* support node os override

* fix multimaster test and os override

* make image available for online test

* cleanup yumrepo and registry post-test
Co-authored-by: Jerry Jackson <jerry@weave.works>
Co-authored-by: Jerry R. Jackson <jrryjcksn@gmail.com>

Relocate reconcilers from wksctl to CAPEI (cluster-api-provider-existinginfra)

Signed-off-by: Dennis Marttinen <dennis@weave.works>

Signed-off-by: Dennis Marttinen <dennis@weave.works>

Signed-off-by: Dennis Marttinen <dennis@weave.works>

Signed-off-by: Dennis Marttinen <dennis@weave.works>

Signed-off-by: Dennis Marttinen <dennis@weave.works>

Signed-off-by: Dennis Marttinen <dennis@weave.works>

Co-authored-by: Jerry Jackson <jerry@weave.works>

Use runtime-controller client interface

Signed-off-by: Dennis Marttinen <dennis@weave.works>

Signed-off-by: Dennis Marttinen <dennis@weave.works>

Signed-off-by: Dennis Marttinen <dennis@weave.works>

Remove bogus taint check in controller

Fetch machine identity fewer times

Code was checking the `Value` of the taint, whereas the string
"NoSchedule" is an `Effect`.  So it would always be false.

Exclude node being updated from HA check

Use the generic Serializer in libgitops, and fix some API machinery nits