Decide on versioning strategy

[robobario]

We could follow semver, but maybe a minor bump is justified when we bump the flink version.

Or we could be releasing artifacts including a flink version since it's quite fundamental. Like maybe we should be releasing an x.y.z-1.20 tag. So the project maven version could follow the semver of the flink-sql job API and we use the flink version tag to show what flink version we're building against.

[tomncooper]

I think a minor version bump on the semver for a flink change is ok.

I think we should probably just leave it as semver and have a compatibility matrix in the README. It is likely that in future more compatibility variables turn up (Kafka, Kubernetes, etc) so keeping them in a table rather than the main version makes sense.