Derive tester/testee based on block hash behind checkpoints

[msgmaxim]

Nodes on different forks will derive different tester/testee pairs if we use the latest block hash. If these nodes test one another, this could result in test being recorded as failed, which would be incorrect.

[jagerman]

Test heights are now derived as block_height_ - TEST_BLOCKS_BUFFER (the latter == 4), and although that often isn't before a checkpoint, it still causes problems because we select random indices into the current SN list: this ends up causing an ocassional "we are NOT the testee" warning, presumably because of new registrations/decommissions/activations resulting in different selection lists between tester and testee.