Refactor math CAT to use downex column of corpus

[zwatson2001]

This PR makes changes to the math timeline so that it is consistent with the other new downex tasks, using the downex column of the corpus for all downex trials rather than relying on a negative difficulty value for instructions and block_index=3 for test trials.

[asengupta3]

Do you think at some point we can make this generic enough where this is config driven and hard coded like right now? Ideally, we will be able to switch the blocks around, along with filtering of ids etc via some external config which then this code reads and builds the task. What do you think?

[zwatson2001]

This kind've a special case, because the math cat is organized into 3 blocks but the non-CAT version has more blocks. Each block has a generic instruction at the beginning saying "Here's a new kind of question", but because the corpus has one for each block in the non-CAT version, the CAT has to filter some of them out since it has fewer blocks. We could maybe add a corpus column cat_only so that it isn't hard coded like this? I'm not sure what the best solution is

[asengupta3]

LGTM! Just one future suggestion probably worth discussing at some point.