Use X-Ray to analyze complete RECAP corpus

[mlissner]

I ran an AWS batch job on this so far. It ran overnight and produced about a gig of results for the entire RECAP corpus. Some of the results are timeouts that we could do a better job catching and retrying, many are false positives, but there are true positives too.

I'll use this issue as a meta-issue to group together things that need fixing.

[mlissner]

• Text on top of rectangles is visible but creates false positive