Consider strategy for testing upgrades across multiple versions

[EvanBoyle]

We recently added tests for GCP v6 -> v7. We know that a significant number of our customers don't upgrade in a timely manner, so we should consider how we continue to accrue testing across versions in future upgrades. For instance, can we have tests for both v7->v8 and v6->v8?

[t0yv0]

Thank you for filing this. Without making any changes to v0.0.2 of this library, one can add several distinct tests with a different baseline version, and that should work in principle to test v7->latest, v6->latest which would work if latest=v8.

That said it seems that it should be easy to support testing arbitrary X->Y upgrades and having a documented way to do that, so if one wanted to test v6->v7, v7->v8 or v7->v8 that should be doable. We might need to make some small tweaks to enable this.

[t0yv0]

@guineveresaenger brings up the need to figure out automation to support this feature as well. Upon a new release of a provider, we need to add (or move) a new baseline, and to support test accelerators we need to perform recording of the provider's behavior at the new baseline which can be time-consuming. Ideally this chore is automated and happens in a way that is non-obtrusive for developers.