Fix off-by-one data

[guidopetri]

A lot of the data was run with an off-by-one error present in the positions/evals/clocks taken from the cloud or from a Lichess-side computer analysis. To fix:

• truncate data
• re-run entire pipeline
• re-run WP model
• re-deploy WP model

The most annoying part will be the evals, which take a long time to run. As a result I'll probably only do this once #55 is finished - that way that can run for super long and I don't have to babysit the pipeline.

[guidopetri]

It might also be worth it to rerun everything after #41 and #29 are done, so that I can get those evals easily too.