specify convolutional width per convolutional layer

[StevenSong]

What
let user specify the convolutional kernel dimensions for each convolutional layer instead of using the same fixed dimensions across all conv layers

Why
some experiments may benefit from varying convolutional kernel dimensions, also I want to be able to try out architecture from this paper (https://www.nature.com/articles/s41591-020-0870-z) which uses conv layers of 16, 32, 64, 128, 256, 512 and conv (1D) widths of 80, 40, 20, 10, 5, 3

How
modify make_multimodal_multitask_model function to take list of conv dimensions and change command line args to take lists for conv x, y, z

Acceptance Criteria
Model architecture that has varied convolutional kernel dimensions per convolutional layer