zombie unit with failed unit assignment cannot be removed

This case seems similar to LP#1643430 but our controller has been upgraded to 2.1.1 and the problem is still present, so this may be a new bug.

We have a model in the following condition:

model:
  name: stg-ols-scasnap
  controller: prodstack-is
  cloud: prodstack-45
  region: bootstack-ps45
  version: 2.0.2
machines: {}
applications:
  sca-postgresql:
    charm: local:trusty/postgresql-0
    series: trusty
    os: ubuntu
    charm-origin: local
    charm-name: postgresql
    charm-rev: 0
    exposed: false
    life: dying
    application-status:
      current: error
      message: 'cannot assign unit "sca-postgresql/0" to machine: cannot assign unit
        "sca-postgresql/0" to new machine or container: cannot assign unit "sca-postgresql/0"
        to new machine: unit is not alive'
      since: 16 Mar 2017 04:11:24Z
    units:
      sca-postgresql/0:
        workload-status:
          current: error
          message: 'cannot assign unit "sca-postgresql/0" to machine: cannot assign
            unit "sca-postgresql/0" to new machine or container: cannot assign unit
            "sca-postgresql/0" to new machine: unit is not alive'
          since: 16 Mar 2017 04:11:24Z
        juju-status:
          current: lost
          message: agent is not communicating with the server
          since: 16 Mar 2017 04:11:24Z

and attempts to remove the unit silently fail. Because unit cannot be removed, we cannot remove the application, and so therefore we cannot remove the model.

Attempts to upgrade the model to 2.1.1 fail:

ERROR some agents have not upgraded to the current model version 2.0.2: unit-sca-postgresql-0

but since there are no live agents in the model anyway, is this even relevant?

Other IS folks: our RT for this is #100760.