[stats] Migrate bayesian engine to gaussian effect model; unlock priors + CUPED

[lukesonnet]

THIS PR IS DESCRIBED IN THE SUMMARY BELOW AND THE FOLLOWING TWO CHILDREN WHICH HAVE BEEN MERGED IN TO IT: #2407 #2409

Summary

Removes our gaussian mean and binomial proportion bayesian models and replaces them with the gaussian effect model (already in use by quantile metrics).

This internal document covers many of the details for what and why: https://www.notion.so/growthbook/Bayesian-Model-Transition-567387e78f4444ba9718fa9d955eab1b

Documentation for this model that will be merged eventually can be found here: https://www.notion.so/growthbook/Statistical-Details-ad44401a63ea402a9aef2944541e56ba

Also tweaks risk to not always be relative when coming back from the stats engine; it now depends on the difference type.

Testing

Already confirmed differences with existing models in above notion doc, will link python notebook at a later date.

TODO:

• Test prior rescaling for absolute differences
• Add unit tests for posterior distributions (this is covered by the existing quantile test)

Follow-up PRs:

• [stats] Add ability to specify effect priors #2407
• [stats] Add CUPED to Bayesian engine #2409

[github-actions]

Your preview environment pr-2406-bttf has been deployed.

Preview environment endpoints are available at:

• https://growthbook-3000-pr-2406-bttf.growthbook.okteto.dev
• https://growthbook-3100-pr-2406-bttf.growthbook.okteto.dev

[github-actions]

Your preview environment pr-2406-bttf has been deployed.

Preview environment endpoints are available at:

• https://growthbook-3000-pr-2406-bttf.growthbook.okteto.dev
• https://growthbook-3100-pr-2406-bttf.growthbook.okteto.dev

[lukebrawleysmith]

lgtm