[Product] Explain recommendations and capture high-value repository actions

## Problem

The feed card does not clearly explain why a repository was recommended and does not expose several high-value actions needed by both users and the ranking system. The README popup presents repository text, but opening it is not logged as a distinct signal.

A recommendation product should distinguish semantic relevance, social proof, trending discovery, and exploration rather than presenting every item as an unexplained model output.

## Proposed change

- Show a concise `Why you're seeing this` reason based on safe, user-facing features.
- Add an explicit Open GitHub action and record it separately from README/details opens.
- Render README markdown safely rather than as undifferentiated plain text.
- Show owner identity/avatar, activity/freshness, license, and repository health where available.
- Add share/follow/not-interested actions and optional not-interested reasons.
- Include the selected board name/category with save events as a consented preference signal.
- Never expose raw internal scores or sensitive inferred attributes.

## Acceptance criteria

- Every card can display a typed recommendation source and at least one human-readable reason.
- README open, GitHub open, save-to-board, share, follow, and explicit dismissal are distinct events.
- Reasons are deterministic from response fields and covered by snapshot/unit tests.
- Accessibility labels and external-link confirmation behavior are implemented.
- Product analytics can compare action rates by recommendation source without relying on raw click-through rate alone.

Start by locating the feed card, README popup, recommendation response fields, and analytics event handlers described in the issue. Read the existing snapshot and unit-test setup before mapping the acceptance criteria to the current UI and events. Done means the distinct actions, typed reasons, safe README rendering, accessibility behavior, and source-level analytics comparisons are covered by tests.

Written by the indexing model from the issue text.