feat: add personas

[woodruffw]

WIP.

This adds the concept of a 'persona' to the determinations set. The thinking here is that personas will model common sensitivity thresholds, allowing zizmor to move from a hard-coded "pedantic" persona to a more general filtering mechanism.

Right now, the personas are:

1. auditor: someone doing secure code reviews, who is insensitive to false positives since everything is being reviewed manually
2. pedantic: someone who is sensitive to false positives but is OK with true positives that are "code smells" rather than actionable weaknesses
3. regular: someone who wants only actionable true positives (or as close to true positives as a given audit can provide)