mining_prioritisetransaction.py checks that prioritisetransaction alone makes getblocktemplate return a new block, by comparing the whole template before and after moving mocktime 10 seconds forward. But curtime comes from the clock, so the two templates always differ, even if the template wasn't rebuilt.
This PR checks that the deprioritised transaction drops out of the template instead.
If you remove the counter bump in CTxMemPool::PrioritiseTransaction (the fix this check was added for in #10196), the test from master still passes but this one fails.