Marshalling scenarios

[ericvergnaud]

@gueniai
The purpose of the PR is to demonstrate that support for weak typing (a fundamental feature of Python!) is a required feature when storing configs, by providing support for real life scenarios. The concern was raise as part of #189

To that end, this PR:

• patches Installation.py to dynamically support or not support raw typing and weak typing
• provides a series of real-life scenarios, each of which is run in the 3 modes (allowing or not raw and weak typing)

When running the test_marshalling_scenarios.py locally, you'll notice that:

• all 8 scenarios pass when weak typing is allowed
• only 2 scenarios out of 8 pass when only raw typing is allowed
• only 1 scenario out of 8 passes when strict typing is required

I don't think it's realistic to expect that developers get things right upfront, we should therefore support their progress towards strict typing. That's how these features ended up here in the first place (I couldn't figure out a way to specify a recursive type for JSON), so it's very likely to happen again and again.

More importantly, it is quite likely that we will encounter scenarios where the data to be stored comes from a source we don't control, and it just needs to be passed through, see test_dynamic_config_data.

Finally, and critically, I also know from experience that as a team we will encounter orphan data, i.e. data that somehow got there, but the how is lost. Support for weak types is key to rapid resolution.

I am yet to see a scenario where supporting this would cause problems in the future, rather the opposite.

Hopefully this is convincing enough to force merge #189 (the PR is already approved, 'force' is only required because some pre-existing code does not pass the 'no-cheat' check, all other checks are green)

@sundarshankar89 FYI

[github-actions]

✅ 40/40 passed, 2 skipped, 1m55s total

_{Running from acceptance #298}