Prototype SongForge lyric and prompt-pack mode

[WalksWithASwagger]

Context

The user explicitly floated turning the product into a song. The useful first version is a bounded creative mode that turns a transcript or KB cluster into lyrics, spoken-word structure, and music-generation prompt packs with source notes.

Source: docs/WHISPERFORGE-MASTER-PLAN-2026-05-18.md (wf-songforge, Phase 9).

Linear: BC-232.

Acceptance Criteria

• Add a SongForge recipe or mode that extracts themes, motifs, phrases, and emotional arc from a capture plus selected KB context.
• Generate at least three structured outputs: lyric draft, spoken-word variant, and service-agnostic music prompt pack.
• Include source notes or receipts explaining which transcript/KB moments informed key themes.
• Keep generated lyrics original and based on user-provided/source material, not imitation of living artists.
• Add documentation and fixture coverage for the creative workflow.

Tests/Evals

• Credential-free fixture test for output shape and source-note rendering.
• Optional UI smoke if exposed in the recipe picker.

Verification

• make test
• make eval-fixture
• git diff --check

Agent Instructions

Keep the first version text-first and service-agnostic. Use the existing recipe/pipeline/export architecture where possible.

Out of Scope

Do not integrate with a specific music generation service or generate audio in this issue.