We need to find ways to ensure that patch/non-breaking releases are compatible across a particular release.

We could do this by enabling multi-version testing in our E2E tests for all release branches (v0.34.x, v0.37.x, etc.) in our nightly tests. The aim here would be to run tests where the tip of the release branch is built and run against the latest release from that branch (e.g. v0.34.x would be built and run on some nodes, and v0.34.24 would be run on others, if v0.34.24 were the latest v0.34 release).

The proportion of "old" to "new" nodes should be controllable somehow, even if it's just set in a constant in the code.