Stop clearing all watches in the mock trigger engine.

[jtibshirani]

Our watcher integration tests use ScheduleTriggerEngineMock to give more
control over when watches are triggered. This mock trigger engine maintains a
reference to all registered watches. We've seen a few instances of tests
failing because ScheduleTriggerEngineMock could no longer find a watch that
was previously added (for example #35503, #37882, and #39306)

The issue seems to be that after the tests starts, there can be a change in
local shard routing, which then causes the watch service to be loaded. This in
turn calls ScheduleTriggerEngineMock#pauseExecution, which clears all watches
that this mock object holds. Subsequent calls to
ScheduleTriggerEngineMock#trigger then fail to find the requested watches.

This PR proposes to stop clearing the watches in
ScheduleTriggerEngineMock#pauseExecution. Usually, pausing execution on a
trigger engine will stop scheduled watches from running. However, the mock
trigger engine does not run watches on a schedule, and instead allows them to
be triggered manually. So arguably, no action should be taken in
pauseExecution, and in particular it doesn't make sense to clear the list of
registered watches.

Addresses #35503 and #39306.

[elasticmachine]

Pinging @elastic/es-core-features

[hub-cap]

Good catch! I ran this for a few hours and did not see any errors crop up because of the change.

[jtibshirani]

Thanks @hub-cap for the review and the extra testing!