I realized I made a mistake above, so here's the fixed resulting code:
private double ExecuteSingleIteration()
{
m_Setup?.Invoke();

if (m_GC) StartGCRecorder();

var executionTime = m_Watch.Split();
m_Action.Invoke();
executionTime = m_Watch.Split() - executionTime;

if (m_GC) EndGCRecorderAndMeasure(1);

m_Cleanup?.Invoke();
return executionTime;
}

private double ExecuteForIterations(int iterations)
{
var executionTime = 0.0D;
for (var i = 0; i < iterations; i++)
{
executionTime += ExecuteSingleIteration();
}
return executionTime;
}
Also, be warned that likely why StartGCRecorder is not actively ran before every iteration by default is that it triggers GC.Collect - you'd need to patch more if you run many iterations (I did 512 iterations for 256 measurements, one run takes a whole minute instead of 1ms).

The fix for single frame tests is pretty straightforward(using 3.0.3 as reference):
1) MethodMeasurement.cs, ExecuteForIterations - move StartGCRecorder and EndGCRecorderAndMeasure inside the else branch
2) MethodMeasurement.cs, ExecuteForIterations - replace ExecuteActionWithCleanupSetup call with ExecuteSingleIteration

That'll fix the GC tracking, but there's also an issues with SampleGroup tracking - they'll also include SetUp invocation