Train 1D convolution as uncertain fusion alsogtihm

[deebuls]

we currently have

1. propagated tensor
2. current tensor

If we stack them and run a 1D convolution through them and run through the same loss function we can get a neural fusion.

@deebuls Deadline July 15