Fixes to ensure Delete phase is always handled and prevent deploys from being stuck in a Savepointing phase

[glaksh100]

Problem

• Deploy starts — Savepoint is triggered and triggerID registered
• GetSavepointStatus call keeps failing for whatever reason
• Retries are exhausted for GetSavepointStatus, however we have no real way to progress after this. The app gets stuck in a Savepointing Phase.
• User tries to force delete the application. However, with retry counts not being reset and a non-nil LastSeenError, the operator never handles the Deleting phase (i.e. isTimeToHandlePhase always evaluates to false).

The PR fixes these problems in the following way:

• Reduces retries for the GetSavepointStatus call from 20 --> 3 to reduce the time to detect savepoint failures (esp. 404s around savepoints not being found for a triggerID). I considered making this non-retryable, but I'm guessing there's a possibility that the savepoint status for a triggerID may not become immediately available.
• When retries for GetSavepointStatus are exhausted, we cannot reliably rollback at this point, as we may not have a running job or a successful savepoint. So we allow for the remainder of the savepointing block to complete i.e. try to recover from an externalized checkpoint.
• When a Delete is attempted, always handle the Deleting phase irrespective of what error/retry state the application is in.

[glaksh100]

@mwylde can you PTAL?

[mwylde]

LGTM. I think this is a great start, but we should continue thinking about ways to make this pathway more robust.