Optimize column placement for PIL backend

[qwang98]

Currently, each signal has its own column, which does not save the number of columns the most efficiently. Optimization can go in two stages:

1. Have # of columns == maximum # of signals for each step type. For example, step type 1 has signals a and b, step type 2 has signals a, c, and d, so max(2, 3) = 3, and we will have 3 columns, with some different signals sharing the same column across different steps. In our current implementation, we would have 4 columns corresponding to 4 signals, which is not the most column saving method. This should be quite doable.
2. Further optimize and reduce the # of columns by using multiple rows for each step type. However, I'd point out that PIL only supports rotation of one (similar to our next for forward signals), and I'm not sure if this will work for step instances with multiple rows though. Therefore, this optimization would require some research to see if there's a possible implementation.