deadlock mitigation redux

[jessepeterson]

While I thought we addressed this in #75 it was not. There are two delete operations: one for the commands and one for the step. I believe we mitigated the command deletion deadlock with #75 but the step deletion deadlock remained. To help with this we:

• rolled the row locks into the step id command retrieval query vs. a separate lock query.
  • also using joins to extend the lock, even though we aren't using that joined info
• narrowed the step deletion to just be related to the relevant command step, rather than all steps which should already be locked due to the above.

Hopefully this gets us where we need to be. Both Go unit tests and mdmb tests were done to verify operation.