What backward compatibility testing is done to ensure the latest release passes previous releases' versions of the functional tests?
There is a feature_backwards_compatibility.py functional test which spins up functional test nodes from each of the previous releases to do testing between them at the same time.
Once you have built the latest version of Bitcoin Core you can download the previous release binaries using:
test/get_previous_releases.py -b v0.19.1 v0.18.1 v0.17.2 v0.16.3 v0.15.2
You can then run the feature_backwards_compatibility.py
test individually or through the test runner.
e.g. test/functional/test_runner.py feature_backwards_compatibility.py
This test is updated with each new release to add a functional test node from the latest release to functional test nodes from previous releases.
At the time of writing (December 2020) there is an open pull request from Sjors Provoost to add v0.20.1 to the backwards compatibility test.
And conversely what backward compatibility testing is done to ensure previous releases pass the current unit and functional tests?
I don't know at this stage if this in any way a priority (or for that matter whether it should be). It is certainly more resource intensive to build entire previous releases and then test them rather than merely spin up functional test nodes from previous releases. (I will update if I learn more.)