Check that single stepping the sig-return instruction, returning to the trapped instruction in the callee, where that trapped instruction has been replaced by a breakpoint, triggers bith the code and instruction observers, and doesn't go one instruction too far.
Most of this is tested by the testStepping* variants in TestBreakpoints, counters on both Instruction and Code observers are observed. An explicit test could be added for "doesn't go one instruction too far".
Checking is is very important, kernel and code bugs if encountered in the past have exibited exactly those edge case problems - executing the instruction, continuing, ...