Implement new improved retry logic

[r4victor]

Closes #1200

This PR introduces retry property to profiles and run configurations as a replacement for retry_policy. It allows specifying different events that should be handled with retry. For example, users may retry provisioning the compute if there is no capacity:

type: dev-environment
ide: vscode
retry:
  on_events: [no-capacity]
  duration: 5m

Supported events are no-capacity, interruption, and error.

The duration is now calculated as a run age for no-capacity and as a time passed since the last interruption and error for interruption and error events. For example, when specifying

retry:
  on_events: [interruption]
  duration: 5m

dstack will try provisioning a new instance for 5m since the interruption. If there is no capacity for >5m, the run fails.

TODO:

• Update CLI run attach mode to support retrying runs – currently CLI exits after run errors.
• Update CLI retry syntax to support specifying retry events.