Incorrect run_left behavior #126
Description
Let's say you press run_left on Block B and it reruns an edited dependency Block A. It should then run blocks between Block A and Block B even if they were already run and unedited (it currently does not)
Example:
On the MNIST example, if I run_left the Training block:
And if I modify the load dataset block by replacing x_train by None
And then if I run_left the Training block again
Pyflow ran Load dataset (setting x_train to None) and ran Training (returning an error because x_train is None)
It should have ran Load dataset and then ran Normalize dataset and stop because normalizing None would have returned an error