[YBazanPro]

What changed

Follow-up to spec 015, from first use of the new chat.

• Refusals explain themselves. When retrieval finds nothing above the minimum score, the request trace, the retrieval.completed event and the activity panel now carry the best candidate score, e.g. "nothing reached the minimum score of 0.55; the closest passage scored 0.48". Before, a refused question showed no detail. New nullable best_score trace column (migration TraceBestScore); SearchResult/SearchRun gain an optional BestScore.
• Compact secondary controls. Helpful / Not helpful / Report issue, Activity, scope and New conversation buttons are small and muted; the Activity toggle shows a chevron.
• Delete buttons are icons. A small inline SVG trash icon (TrashIcon) replaces the "Delete" text. The Fluent icons package was not reintroduced, to keep the client download small.
• Panels are identifiable. Conversation list, thread, scope bar and composer are bordered cards with small titles ("Conversations", "Search scope").

Reviewer notes

• Why the question in the screenshot was refused: the dev trace shows NoEvidence with no passage above 0.55, so the chat model was never called; a general question ("what is the book about?") scores low against any single passage. This PR makes that visible; it does not change thresholds or add overview retrieval.
• The same trace shows retrieval_ms of 8190 for that question; not investigated here.
• Visual result not checked in a browser; the running app was an older build.

Verification

dotnet test --solution RagAssistant.slnx -c Release: 1043 passed; dotnet format --verify-no-changes clean. New tests cover the best score in the activity endpoint, the stream event and the panel text.

🤖 Generated with Claude Code