Pin MarkupSafe < 2.0.0 for py35 support

[ajkavanagh]

MarkupSafe 2.0.0+ (required for Jinja2) removed py35 compatability, so
we need to pin it for charms to work on xenial.

Closes-bug: #196

[ajkavanagh]

Looks like the juju bit failed; testing locally to work out why.

[ajkavanagh]

I re-ran the failing func test locally: with the following output:

2021-05-13 10:58:31 [INFO] Ran 1 test in 0.007s
2021-05-13 10:58:31 [INFO] OK
2021-05-13 10:58:31 [INFO] Metadata: {}
Events:
  Deploy Bundle:
    PCT Of Run Time: 4
    Start: 1620903296.0478487
    Finish: 1620903302.9090686
    Elapsed Time: 6.861219882965088
  Prepare Environment:
    PCT Of Run Time: 2
    Start: 1620903293.3278556
    Finish: 1620903296.0477326
    Elapsed Time: 2.719877004623413
  Test zaza.charm_tests.noop.tests.NoopTest:
    PCT Of Run Time: 1
    Start: 1620903511.783317
    Finish: 1620903511.7983708
    Elapsed Time: 0.015053749084472656
  Wait for Deployment:
    PCT Of Run Time: 96
    Start: 1620903302.9091117
    Finish: 1620903511.6250138
    Elapsed Time: 208.71590209007263

______________________________________________________ summary ______________________________________________________
  func: commands succeeded
  congratulations :)

[johnsca]

LGTM and I also got a passing manual run. No idea why it's failing on GH; it doesn't seem related to this change, but it is a bit worrisome. I added a commit to make the status and error log steps be run even on error to try to get more info.

[hloeung]

Any updates on landing this PR?

[ajkavanagh]

Ping @johnsca ; please can you take another look and see if we can get this merged? Thanks.

[johnsca]

The functional tests are failing due to consistent provisioning errors that don't seem to happen when running things locally. I'm guessing it has to do with some change with LXD or Juju and the fact that they're run in fresh VMs (it seems like there are two instances provisioning with the same images as the ones that are failed but which haven't finished downloading yet, so maybe there's a duplication check race condition).

At any rate, not worth blocking this fix for since we have manual confirmation of the change, but we will need to get it sorted out ASAP so that future PRs don't have the same problem.