Lazily compute row and column roots when repairing the square

[evan-forbes]

When repairing the square, we are currently generating roots of each column and row even if there is nil data in that row or col. This is a problem because because if the underlying tree doesn't know how to handle nil values, then it has no choice but to panic. While there are quite a few ways to fix this, I think the fastest way would be to only compute the root of columns and rows that have no nil values. In the context of RepairDatasquare, this means having the ability to generate roots of single columns and rows.

This must be fixed before #235 and #232 can be completed.

[liamsi]

While there are quite a few ways to fix this, I think the fastest way would be to only compute the root of columns and rows that have no nil values. In the context of RepairDatasquare, this means having the ability to generate roots of single columns and rows.

That sounds good to me. Do you mind drafting a minimal PR (without tests etc) to see how this would look like in code? At least it sounds like a relatively simple change.

[evan-forbes]

I cheated and wrote the code before the issue 😄 #18