Implement cluster rebalance rollback and cleanup

[whitehawk]

Implement cluster rebalance rollback and cleanup

This patch adds:

1. Full rollback (via '-r' flag) capability for the rebalance flow.
2. Error checking for the steps marked as errored due to the interruption of the
   tool execution (during both normal and rollback flows). Errored steps can be
   retried, rolled back individually, or cancelled.
3. Cleanup capability when the execution was interrupted during rebalance.

In order to facilitate the capabilities above:

1. Added SegmentStatus command class to check the status of a running segment
   via 'pg_ctl'.
2. Added fault injections inside the 'gprecoverseg' tool for the test purposes.
3. Renamed 'wrap_state_func_with_faults' into 'wrap_func_with_faults' in the
   fault injector, as now it is used not only with the state machine functions.
4. In RebalanceStep class removed rollback-related statuses, as they duplicate
   the statuses of the normal flow. Instead, 'rollback' flag is added to
   distinguish between the steps of the normal and rollback flows.
5. Introduced updates in the main and shrink state machines in order to
   coordinate cleanup and rollback between shrink and rebalance.
6. Removed 'gprecoverseg' invocation from 'check_down_segments()', as now the
   proper error handling is done by the rebalance execution engine itself.
7. Moved 'check_down_segments()' to an earlier point to make FTS probe before
   'check_running_gputils()' (otherwise, in some test cases
   'check_running_gputils()' fails when accessing the cluster).
8. Added '--skip-resource-estimation' option to 'gpmovemirrors'. We use this
   option when moving mirrors, as, when retrying a failed operation, in some cases
   there will be no old datadir yet, and 'gpmovemirrors' will fail at resource
   estimation. So we skip resource estimation at 'gpmovemirrors' side. It is not a
   problem, as we have our own resource estimation.
9. Added new behave tests for rollback and cleanup.
10. Added new steps for behave tests.
11. Removed cases, related to failures during switchovers, from rebalance test
    '3', as these cases are not checked in the newly added test cases differently.
12. Removed cases with delay from rebalance test '3', as now such scenarios are
    tested with the help of fault injections in the 'gprecoverseg' at particular
    points of interest.