Unfortunately, after doing some more extensive testing and fixes here and there, it turns out that Ansible 2.0.x and 2.1.x have a number of deficiencies that make them either tedious to work with, or even dangerous in terms of catching errors.

In particular, the following issues can be observed:
* Playbook runs take way longer than 1.9.x. Something along the lines of 5-6 times longer.
* If specifying an unused tag, Ansible will merrily continue on its way executing stuff. With Ansible 1.9.x this would've produced an error. This is apparently a "known issue" (sic) in Ansible 2.
* If a task is notifying a handler, and the handler of that name does not exist, this does not trigger an error. I.e. if you have miss-typed the handler name, you'll never notice it until you realise that something did not get updated.
* Dynamic names for handlers work worse than before. Not sure why, but there's just something (even more) borked in the way inlined vars are expanded. The tasks get straight-out skipped (this is a bit related with the previous point).

For now it looks like Ansible 2 is a bit more rough on the edges than it should be, and it is better to stick to 1.9.x. Keeping forward-compatibility is fine, but certain things will need to be rolled-back. Mainly:
* Documentation should refer to 1.9.x as officially tested, with 2.0.x and 2.1.x kind of supported, but one should not use them with Majic Ansible Roles for now.
* Usage instructions should be based on Ansible 1.9.x instead of Ansible 2.1.x.

A full run-through of the testsite and usage instructions should be performed as well, just to be on the safe side.