Fix tsh/common test timeout by sharing kube cluster fixtures

[jakealti]

The tests in tool/tsh/common were timing out in CI.

When a test changes, CI runs it 100 times to check for flakiness.
That job has a 10 minute limit.
TestKube and TestKubeLogin start three Teleport clusters between them,
and they did it on every one of the 100 runs.
So once a test with this setup was added, the job ran out of time.

Now the clusters are started once and reused.
There are two of them, one for each proxy mode the tests cover.

TestKube + TestKubeLogin on my machine, with the same flags CI uses
(-race -shuffle on):

	before	after
-count 1	20.5s	16.7s
-count 10	139.5s	24.3s

Before, each extra run added about 13 seconds. Now it adds less than 1.
That takes 100 runs from roughly 20 minutes down to under 2.

Only tests changed. No change in behavior.