Safeguard unit test threading

[Cliabhach]

Ideally, this will remove some flakiness from our tests in AppDataManagerTest.kt.

Said uncertainty was introduced by 9d41f30.

This PR also reduces the memory usage of the JVM build process (advice derived from https://discuss.circleci.com/t/circleci-2-0-android-with-kotlin/17531/2 ) and switches the version of the Android Gradle Plugin in use to the latest stable release (v3.2.1).

[Cliabhach]

@caguilar187 Let me know how you feel about that latest change.

[Cliabhach]

As requested offline by @caguilar187, I am running the full CircleCI integration multiple times against commit 782ac6e. If it passes 70+% of these runs, we will consider this a sufficiently-worthwhile improvement and thus worthy of being merged in.

[Cliabhach]

We're looking at 50% success after 8 runs.....breakdown of failures as follows:

2 failures due to the daemon disappearing (tasks :welcome:packageDebugResources, :access_card:lint)
1 failure due to integer number too large: 7f0ff007c (task :welcome:kaptGenerateStubsDebugKotlin)
1 failure due to D8: Cannot allocate memory (task :app:transformDexArchiveWithExternalLibsDexMergerForDebug)

I personally believe all but the 'integer number too large' failure are affected by the memory allocated during compilation, especially of those modules running kapt. If we reduced the associated load by e.g. consolidating modules (as in #299) or re-evaluating the complexity of some of our database objects, I'm sure the failure rate would decrease as well. I'll now take the opportunity to rebase this branch on the latest dev.

@caguilar187 your thoughts?

[Cliabhach]

I'll rebase on dev and try something different: the latest pre-release AGP version (3.4.0a2).

[Cliabhach]

We've reached 7/8 successes for this last commit.