Introduce `runtime.ForMatrixVersion`

[ShadowJonathan]

Related to #335, and as a related alternative to what I was trying to do in #242;

Introduce runtime.ForMatrixVersion(major, minor uint8), which will fetch /client/versions and compare if the given version is compatible with any version given.

If not, skip the test.

This'd make it possible for complement to skip whole hosts of tests automatically if the server does not broadcast support for it.

It would also self-document the feature prescribing itself only for/after a certain matrix version, which would help with sorting through them at a later time.

In the future, it could also be possible to do the same for MSCs (via `unstable_features matching, and such)

[clokep]

Is the assumption here that runtime.ForMatrixVersion(1, 4) would mean that anything >= 1.4 would be OK?

I'm unsure if that's valid.

[ShadowJonathan]

>=1.4,<2.0, more or less.

Do you think specifying an upper bound (for when a feature is deprecated) is needed?

[clokep]

Do you think specifying an upper bound (for when a feature is deprecated) is needed?

For what we have today, probably not. I guess we could always handle that in the future.

[kegsay]

Closing this. We want to move away from Complement dictating what tests are run. Whilst I like the idea of having Complement self-configure itself, it could also be a bit of a nightmare. If /versions omits a previously working version, tests will be skipped and nothing will fail. This feels like a bad failure mode which can easily be avoided by just stating what versions you support at test execution time.