This is an alternative to #8316 for fixing #7148 (comment).
The basic problem with the test that is failing is that we don't have a way to know if the next poisson relay time has occurred, if we're trying to test that nothing is inv'ed. This fix adds something that we expect to be inv'ed, and the test now assumes that anything else that would be relayed would have been relayed at the same time.
It's unfortunate that this is still somewhat dependent on how transactions are batched for announcement, but this at least improves the situation so that the test should now reliably succeed on master. And given that the implementation of feefilter is highly dependent on the transaction relay code anyway, it's not clear how much of a separation we could reasonably expect to achieve in the test.
