Handle proportions for RA and uplift metrics

[lukesonnet]

Features and Changes

• Refines regression adjustment approach for proportions. Required two changes: handling compute_theta for proportion statistics and using the unadjusted mean for computing the variance of the relative change (uplift). Note this second change affects sample mean tests as well, and should make it more conservative.
• Also handles case where variance is missing in one of the periods and reverts to non-RA (but doesn't alert the front-end to this fact).

Dependencies

n/a

Testing

• Added unit tests
• python notebook showing that coverage and variance reduction are as expected for both prop and non-prop caes: https://colab.research.google.com/drive/1qMlDH8GXjoaGWNKJqo3lsF8DieoqVeZ5?usp=sharing

Screenshots

n/a

[github-actions]

Your preview environment pr-1090-bttf has been deployed.

Preview environment endpoints are available at:

• https://growthbook-3000-pr-1090-bttf.growthbook.okteto.dev
• https://growthbook-3100-pr-1090-bttf.growthbook.okteto.dev

[lukesonnet]

This is the only change that is on an actively used path, and unadjusted_mean should always == mean for non-RA metrics.