Looking at this function by itself, it looks like this should be a reference. But I see the only call site is initializing it in-place. I'm still relatively new to rvalue references, so is this the right place to use one? If not, why?

Ideally this would be declarative similar to -enablebip61: -enableava
But I can see the value in having the help text for ava be grouped together or easily greppable.

Perhaps the only improvement I can offer is -avaenable to keep the tense consistent.
Over time, we can move towards a format of -<group><attribute>

Since we expect ava to respond quickly, isn't a 60 second timeout way larger than expected, even during a failure? From experience, I know waiting a full minute unnecessarily for a test to fail is a burden on debugging. Although not universal, we have reduced timeouts in some tests due to this.

Call it what it is rather than "fake node": "node interface"

Something like this as assert_ava_response would be useful in the test_framework utils IMO, but that can wait until we have more tests like this one.

Give names to the error codes, at least within this test.

These classes should set __slots__ = ()
This may become relevant if functional tests end up sending thousands of ava messages: https://docs.python.org/2.5/ref/slots.html

Ditto re __slots__

neabled => enabled

AvalancheVote expects an uint32_t

Note that this is treated as an int while the node is treating it as an unsigned.
This is not really an issue since you don't use any "meaningful" unsigned values that would change the sign, but it's maybe a good idea to use an int everywhere, or use constants for the -1, 0, 1 values.