feat: provide clearer restriction reasons and keep post-level restrictions post-level (#216)

[Adityakk9031]

Summary

Closes #216

1. Clear, Actionable Restriction Transparency in Under the Hood

As requested in #216, Under the Hood labels now provide creators with high-level, human-readable reasons, affected elements, and explicit recommendation surface impact without exposing internal classifier weights:

• under-the-hood/strato/lib/underTheHoodLabels.strato:
  • Enriched both postLabels and accountLabels with:
    • reason: high-level readable category (e.g., "Sexually suggestive or explicit visual media detected under Adult Content policy", "Automated spam or repetitive distribution pattern detected").
    • restriction: explicit distribution impact (e.g., "Not recommended to non-followers; content warning applied").
    • scope: distinct classification boundary ("Post-level" vs "Account-level").
    • affectedTarget: specific affected element ("Media", "Text", "Link", "Account profile", etc.).
  • Added helper accessors (postLabelReason, postLabelRestriction, postLabelScope, postLabelAffectedTarget, accountLabel...) and exported them in the Strato Library.
• under-the-hood/strato/columns/underTheHoodReport.User.strato:
  • Updated postLabelRow and accountLabelRow to include reason, restriction, scope, and affectedTarget in reportJson.
• under-the-hood/jetfuel/routes/_lib/report.ts:
  • Extended PostLabel and AccountLabel types and JSON parsing for reason, restriction, scope, affectedTarget, and affectedMedia.
• under-the-hood/jetfuel/routes/report.tsx:
  • Updated LabelCard to render scope badges (Post-level in blue, Account-level in amber), high-level reason rows, affected elements (e.g., Affected element: Media or Image 2), and specific restriction surfaces.
• under-the-hood/jetfuel/routes/_lib/mock.ts:
  • Updated mock data to showcase the enriched attributes matching the RFC specification.

---

2. Keep Post-Level Restrictions Focused on the Post

Addressed the cascade where a single post label turns into an account-wide suppression penalty that hinders subsequent compliant content:

• home-mixer/filters/oon_nsfw_simclusters_filter.rs:
  • Out-of-network recommendation dropping is now focused on candidates that actually carry sensitive/NSFW content or media. Compliant non-media posts from the creator are retained rather than being dropped unconditionally across the board.
  • Added comprehensive unit tests verifying filtering behavior for media, labeled, and compliant posts.
• home-mixer/models/candidate.rs:
  • Prevented stamping SAFETY_BIT_AUTHOR_NSFW in Phoenix neural candidate conversion for compliant non-media posts from authors with author-level historical flags, ensuring clean subsequent content receives fair candidate ranking.
  • Added unit tests for to_recsys_candidate() verifying safety_label_mask on compliant candidates.
• safety-label-user-agg/postToUserLabelRules.strato:
  • Configured mediaOnly: true on the nsfw_high_precision aggregation rule so that non-media posts are not counted toward visual media aggregation.

---

Verification

• Unit tests added in home-mixer/filters/oon_nsfw_simclusters_filter.rs and home-mixer/models/candidate.rs.
• Tested and validated TypeScript interfaces and parsing logic in under-the-hood/jetfuel/.