1. In order to aid future features its useful to split
split apart the linear_flow and have a good base
class that is not connected to a ordered or linear flow.
2. Have the graph flow inherit from the linear flow since
it is performing a linear run using a topological ordering
of the tasks.

Change-Id: I287cd4a666caf09ca3a4cdf1d53814e9c6feb765

Allow other flow types easier access to these helper
classes by moving them to the shared utils file.

Change-Id: I001754efc25b025e8070493d5861ffaedacafc7b

The test pip ordering was resulting in the pep8 rules not really
running so this fixes that up and the resulting errors so we can
get a clean run.

Change-Id: Icc27be05cd513c4f04135c7e768007dbf9d07b10

Change-Id: I5ff975bb5ded48a84341efa4ebfdae4bf430ab60

Instead of only adding the task to the rollback
accumulator after it has ran it is useful to add
it to the accumulator before it runs so that if
said task fails while running it can get a chance
at undoing anything it might have done.

Change-Id: I5c9cf2fd194e3b42639d84a0d9dfed3a3b5a797d

In cases where it is not desired to reset the whole state
of the workflow, but only to soft_reset it so that it can
be easily resumed from a interrupted state (for example).
This change allows for that to happen, which involves also
keeping track of the last task that ran so when resumed via
soft_reset that task can be the one started off from (instead
of starting from the start of all tasks).

Change-Id: I034bd6af6445e3da52356328605368319c4ff6f9

Previously we seemed to have forgotten to notify the listener
when a task errored and allowed said listeners to get the reason
why the task failed so fix that and allow for the exception to
be returned back on result fetching and rethrown.