Allow longer name for AWS/Azure cluster and refactoring

[Michaelvll]

This PR

1. fixes the first part of the TODOs in Gracefully handle long cluster names & for spot, show FAILED_CONTROLLER instead #963, to unblock the user from using a long name for spot jobs on AWS.
2. moves the cluster name checking into each Cloud.
3. refactors the cloud.supports to cloud.check_features_are_supported and change the boolean return value to exception raising.
4. optimizes the failover for InvalidClusterNameError and ClusterUserIdentityError to directly move to the next cloud, as those errors apply to the whole cloud.
5. fixes minor UX issues for failover:
   Previously:
   
   Now:
   

Tested (run the relevant ones):

• Any manual or new tests for this PR (please specify below)
  • sky launch -c i-am-using-a-very-long-name-for-the-aws-cluster-it-should-be-more-than-63-characters-and-hopefully-it-works-this-sentence-has-135-chars --cloud aws
  • AWS 244 characters: sky launch -c i-am-using-a-very-long-name-for-the-aws-cluster-it-should-be-more-than-63-characters-and-hopefully-it-works-by-checking-the-document-of-aws-we-found-that-the-cluster-name-length-should-be-less-than-250-due-to-the-tag-limit-this-sentence-has-246-c --num-nodes 2 --cloud aws -t t3.micro (and 245 does not work)
  • GCP (didn't change the original tested limit)
  • Azure 42 characters: sky launch -c i-am-using-a-very-long-name-for-the-aws-cluste --cloud azure --region southcentralus --num-nodes 2
  • Lambda 64 characters: sky launch -c i-am-using-a-very-long-name-for-the-aws-cluster-it-should-be-mor --cloud lambda works
  • sky launch -c i-am-using-a-very-long-name-for-the-azure-cluster-it-should-be-more-than-63-characters-and-hopefully-it-works-this-sentence-has-136-chars --cloud azure: correctly fails for the Azure
  • sky launch -c i-am-using-a-very-long-name-for-the-aws-cluster-it-should-be-more-than-63-characters-and-hopefully-it-works-this-sentence-has-135-chars --gpus A100:8, it successfully and quickly failover through Lambda/GCP/Azure and AWS (skip the first three clouds with a single try and try all the regions on AWS).

[concretevitamin]

Thanks @Michaelvll, and nice tests! Shall we also test a name length close to AWS's char limit?

[Michaelvll]

Shall we also test a name length close to AWS's char limit?

Good catch! Actually, we have to reduce the limit to 246 for AWS because of the limitation of DescribeInstance API.

[concretevitamin]

Thanks @Michaelvll. Just some minor comments and a question about some additional characters added by the autoscaler. May be worth updating PR description with each cloud's at-limit test?

[Michaelvll]

May be worth updating PR description with each cloud's at-limit test?

Great point! I added the tests for each cloud, and seems there are multiple corner cases that are caught by those tests. I fixed the cluster name limit for AWS and Azure. Thanks!