Assumptions about what information is already public

[Shoeboxam]

There are some remaining assumptions we make of the data--

• ~~the number of columns is known~~
• ~~the number of columns is nonzero~~
• the number of rows is nonzero
• ~~column names~~
• the csv is well-formed
• the path to the csv is public
• the existence of a file at the path is public
• ~~the dimensionality (number of axes) of the original dataset is known~~
• the user properly labels private data as private
• the user properly indicates if the first row of the dataset is a header
• for the rust runtime specifically: the data fits in memory (std lib panics upon failed alloc)

In-progress:

• computation in the analysis won't overflow/underflow
• elapsed computation time is not public

While none of these are explicitly published, they can be acquired via side channels

~The number of columns assumption can potentially be alleviated by generating empty string columns. When an empty string column is cast to numerics, the entire column will be imputed.~ done

I'm not sure how to handle the non-zero number of rows assumption. I have adjusted the resize component to only accept N > 0. But not all statistics need to know N. Perhaps components that don't need to know N could take a default parameter that is always output with some probability? This can appear in interesting ways -- computing the median of data filtered by a very specific mask that may or may not match one individual. EDIT: there is now protection against this with is_not_empty property, but we still make the assumption that this property is true in materialize.

[tercer]

We can have assumptions. We just need to make them clear. The zero row could be an edge case we want to think about more.