In validation thread(s), the typical lock order is cs_main -> cs_vNodes.
Avalanche reverses this by trying to lock cs_vNodes -> cs_main, tripping the deadlock
detection.

Following the same pattern as net_processing, we can lock cs_main and assert the lock
is still held within the ForNode() lambda. This ensures the same lock order as other
threads.

cmake -GNinja -DCMAKE_BUILD_TYPE=Debug ..
for i in {1..20}; do ./test/functional/test_runner.py abc-p2p-ava* ; done

There's still some rarer flaky behavior in this test, but none of the failures should
indicate a deadlock in debug.log.