Neglecting the bigger and fundamental issue of stuck HTLCs and the fact that we don't know which onions / partial payments arrived at the destination we should with this approach still in practice be able to send substential amounts in times comparable to current strategies for payment delivery. Very happy and excited to help you (if that is even necessary) to make this happen!
Now that we can find a reasonable approximation to the min cost flow problem in less than a second it seems reasonable to not only include the uncertainty network to payment delivery logic as suggested in #4920 but to also include the piecewise linear approximation of the min cost flow solver and combine this with the maintance of the uncertainty network.
Neglecting the bigger and fundamental issue of stuck HTLCs and the fact that we don't know which onions / partial payments arrived at the destination we should with this approach still in practice be able to send substential amounts in times comparable to current strategies for payment delivery. Very happy and excited to help you (if that is even necessary) to make this happen!