Run View: explain Results with evidence and warnings

[5aumit]

What to build

Complete the expandable Result assessment so a researcher can understand why each dataset ranked
where it did and where evidence is missing.

Blocked by

#6 and #7.

Acceptance criteria

• Every Result displays a compact Requirement Check summary when collapsed.
• Expanding a Result shows all Requirement Checks, status, explanation, Evidence Citation,
  known issues, dataset-specific limitations, and the original dataset link.
• Not verified is visibly and semantically different from Does not meet.
• A Completed with warnings Run keeps run-wide limitations near the Result Overview and
  dataset-specific limitations on the affected cards.
• The completed Results contract retains or joins candidate name, source, and URL rather than
  reconstructing them from Markdown.
• Replay contains realistic evidence, known-issue, and reduced-coverage examples.
• Browser E2E covers expansion, source navigation, top-three, Show all, and visible warnings.

[5aumit]

Scope clarification before implementation

Treat the following as required for this issue:

• The replay fixture must truthfully show which candidate documentation was unavailable.
• A dataset-specific Result Limitation must identify the affected candidate.
• Every displayed Evidence Citation must match retrieved evidence for that same candidate: matching source kind and URL, with an excerpt present in the retrieved document.
• Render only normal web links (https or http) for citations and dataset links.
• If the evaluator returns duplicate, missing, or otherwise invalid ranks, do not silently invent a new ordering. Reject the response through the safe failure path.
• Keep browser interaction coverage for expanding Results, evidence, source links, and Show all. Layout containment itself is tracked separately in Keep Results usable on normal screen sizes #19.

The active worktree remains the starting point, but these guarantees are required before this issue is considered complete.

[5aumit]

Current-code references

• Retrieved evidence is stored on each candidate here: setscout/graph/nodes/gather_evidence.py:20-42.
• Citation fields currently require a citation for a verified check, but do not compare it with the candidate’s retrieved EvidenceDoc: setscout/models/schemas.py:74-114.
• The presentation currently renders a citation URL directly as a link: setscout/presentation.py:169-180.
• Candidate-scoped limitations currently lack a candidate identifier: setscout/runs.py:72-76.
• Duplicate ranks are currently normalized into a fresh ordering before validation: setscout/graph/nodes/evaluator.py:64-106.

These citations identify the current gaps described in this issue.