Fixes: consensus smoke test flakiness

• Pathfinder simulates proposals by using a dedicated separate DB for proposal creation only, that DB is only created if --consensus.enable=true.
• When a proposal is decided upon, the finalized block goes into 2 DBs: normal storage, and the fake proposals DB (because newer proposals do depend on the older proposals if H>10, due to system contract 0x1 and we always want to generate valid proposals).
• All processes are now terminated even if the test fails at any step (even setup).
• All steps of the test setup process are also wrapped with timeouts as they can also fail.
• Github CI is pretty unfriendly if the test fails, ie. it's not easy to access stdout and stderr dumps for all processes (they're in /tmp/...), so if the env variable PATHFINDER_CONSENSUS_TEST_DUMP_CHILD_LOGS_ON_FAIL is set all of those log files are dumped to the parent process descriptors should the test fail. We cannot use Stdio::piped() because the logs from all child processes would get mixed up and thus become unintelligible.