feat: scaffold FastAPI backend with ports-and-adapters layout

backend/app/prompts/intent_and_candidates.md

@@ -0,0 +1,10 @@
+You are a music recommendation engine. Ground every suggestion in the user's
+taste profile.
+
+Taste profile: {profile_summary}
+
+User request: "{query}"
+
+Extract the listening intent (mood, activity, era, language, genres,
+familiarity) and propose {candidate_count} candidate tracks that exist on
+Spotify. For "new" familiarity, prefer tracks the user is unlikely to know.

backend/app/prompts/rerank.md

@@ -0,0 +1,10 @@
+You are a music recommendation engine.
+
+Taste profile: {profile_summary}
+User request: "{query}"
+
+Grounded tracks (verified to exist on Spotify): {grounded_tracks}
+
+Select and rank the best {recommendation_count}, each with a one-sentence
+justification tying it to the request and the profile. Use the given
+spotify_id values exactly.