[core] Refactoring the return value of teardown

[Michaelvll]

Fixes #1649

This is a minor refactoring that avoids using the return value to indicate the failure of the teardown. It also avoids the sky down's progress bar progress when a teardown fails. cc @mraheja

Tested (run the relevant ones):

• All smoke tests: pytest tests/test_smoke.py

[romilbhardwaj]

Thanks for adding this @Michaelvll! What's a good way to test this PR/simulate down failures?

[Michaelvll]

What's a good way to test this PR/simulate down failures?

Sorry for missing the question. One way to test it is to disconnect the internet and sky down test-down*. With the master branch, the progress bar will show 100%, even though multiple error shows up, but the current branch will show 0%.

[Michaelvll]

This is also important to avoid leakage of spot clusters in managed spot, as we didn't actually catch the exceptions when the sky.down fails

skypilot/sky/spot/recovery_strategy.py

Lines 37 to 39 in 5bb759a

	usage_lib.messages.usage.set_internal()
	sky.down(cluster_name)
	return

[concretevitamin]

LGTM, if it's confirmed that the following works: sky down -a (or multiple clusters) and only some of the clusters throw a RuntimeError.

[Michaelvll]

Just tested sky down multiple clusters: some succeeded and some failed, with the correct progress bar.

I am testing again with the smoke tests and will merge it after all tests pass:

• pytest tests/test_smoke.py