Juju landing is now faster (unless it isn't)

[Ian Booth]

Hi all

There have been a few fixes landed by different people to address various
intermittent test failures. These were present before the cut over, but seem to
show up more when the tests are run on EC2 rather than Canonistack.

After a little experimentation, plus switching the EC2 instance type from
m1.xlarge to m3.xlarge, it seems we are able to get the tests running more
reliably with full parallelisation.

Approx timing:
- start ephemeral instance, ready to run tests: 5 minutes
- run tests: 11 minutes

There's still the safety net of running the tests serially if the first run
fails. But hopefully landing pull requests will be faster overall.

We'll monitor the situation and make adjustments as necessary. I expect we'll
still see some intermittent failures with parallelisation due to our ongoing
mongo / repliaset test flakiness. But progress is being made...