Implement reschedule_on_lost to take tighter control over rescheduling

[DominicLavery]

This set of changes are intended to become the forked version used in production until the changes can hopefully make it upstream.

It implements this idea: hashicorp#10366 The main change is to prevent rescheduling tasks when their client becomes lost. e.g. the agent crashes

Core points:

• Tasks with RescheduleOnLost set to false do not get rescheduled when lost
• RescheduleOnLost defaults to true and is optional for backwards compatibility
• Lost allocations (and their jobs) will not be GC'd if reschedule_on_lost is set to false. This will need to be done manually or some other means to prevent them building up

[DominicLavery]

Closing rather than merging this, as will use this branch as our deployment branch

[github-actions]

I'm going to lock this pull request because it has been closed for 120 days ⏳. This helps our maintainers find and focus on the active contributions.
If you have found a problem that seems related to this change, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.