Update Dockerfiles using golang images to Go 1.16.13

Prior to this commit the Dockerfiles for windows images relied on an
older version of the official golang images.

This PR updates the windows images to use Go 1.16.13.

Update Dockerfiles using golang images to Go 1.16.13

Prior to this commit the Dockerfiles for windows images relied on an
older version of the official golang images.

This PR updates the windows images to use Go 1.16.13.

Update Dockerfiles using golang images to Go 1.16.13

Prior to this commit the Dockerfiles for windows images relied on an
older version of the official golang images.

This PR updates the windows images to use Go 1.16.13.

Fix test panic: ensure annotations map is not nil

Update Dockerfiles using golang images to Go 1.16.13

Prior to this commit the Dockerfiles for windows images relied on an
older version of the official golang images.

This PR updates the windows images to use Go 1.16.13.

Clean up RunsToCompletion interface

The RunsToCompletion interface was introduced so that PipelineRuns and Runs
could conform to a common interface for CloudEvents and subsequently removed.

This commit removes the redundant function PipelineRun.IsTimedOut (redundant with PipelineRun.HasTimedOut).
PipelineRun.HasTimedOut was introduced so that PipelineRun would implement RunsToCompletion.
"HasTimedOut" was kept (as opposed to "IsTimedOut") for the sake of consistency with TaskRun and Run.

It also removes stale references to RunsToCompletion. No functional changes.

Reduce the sleep to 15s

Reduce integration test run times

The TestDuplicatePodTaskRun test creates 50 pods every integration run (25 for
v1beta1, 25 for v1alpha1). On a successful test run in Pipelines' CI cluster
this takes around 49 seconds. Locally in a kind cluster a successful run takes
40 seconds. This commit reduces the number of pods created to 10 per
integration run (5 for v1beta1, 5 for v1alpha1). This drops the run time to
anything from 10 seconds locally to 21 seconds for the first run in our CI
(v1alpha1) and 36 seconds for the second run (v1beta1).

TestStartTime takes 73 seconds in our CI and includes multiple 10 second
sleeps. Reducing these to 2 seconds drops the test run time to 21 seconds.

TestSidecarTaskSupport is sped up by running in parallel and exiting more
quickly when the taskrun reconciler attempts to kill the sidecar containers.
Log spam from this test is also reduced by sleeping inside an otherwise tight
while loop.

Add reference to `Task` authoring recommendations

We have `Task` authoring recommendations in the [Tekton Catalog][recommendations].

As described in tektoncd#3948, users
need these recommendations but they are not easily discoverable.

Therefore, adding a reference to these recommendations in `Tasks`
documentation to make it more discoverable.

Closes tektoncd#3948

[recommendations]: https://github.com/tektoncd/catalog/blob/main/recommendations.md

Fix panic when pending pipelinerun is failed

PipelineRuns that are created with status PipelineRunPending
can be placed into a failed state before their execution begins.
For example: a third-party controller may be watching for pending
PipelineRuns to perform some checks on them prior to execution
beginning. If those checks fail the controller might choose to
set the PipelineRun status to failed with a relevant error message
indicating which check failed and why.

Prior to this commit when a pending PR failed our metrics code
could panic because the PR's StartTime is nil.

This commit adds a guard to the metrics code to ensure that StartTime
is not nil before computing the PR's duration. If it is nil then
we assume the duration is 0. A unit test confirming this behaviour
has been added as well.

(cherry picked from commit 0299c6c)