Dead code / config cleanup

[dkirby-ms]

Summary

Several artifacts are defined but unreachable and should be removed or connected once the real paths land: the orphaned AI Foundry references (if external endpoint is chosen instead), any remaining AZURE_AI_* traces, and the unreachable narration-type branches. This is the final sweep so the codebase honestly reflects what is wired.

Scope / Tasks

• If Issue 1 chose an external endpoint over AI Foundry, delete infra/modules/ai-foundry.bicep and its commented references (main.bicep:196-205,241) rather than leaving them dangling.
• Confirm AZURE_AI_* removal landed (Issue 4) and no doc still references them.
• Remove or justify any narration-type branches still unreachable after Milestone B (backgroundEnrich at NarrationService.ts:135-159 becomes reachable once LLM is configured — verify, don't blindly delete).
• grep sweep for "TODO"/"planned"/"dead" narration comments resolved by Milestone B; update stale comments (e.g. AmbientSystem.ts:5-6).
• Ensure the renamed/repurposed Azure test (Issue 12) leaves no misleading vestige.

Acceptance Criteria

• No commented-out infra modules or dead env vars remain referenced.
• Every defined LLMNarrationType is either reachable at runtime or explicitly documented as intentionally template-only.
• Stale "planned"/"TODO" narration comments are updated to match reality.

Dependencies

Milestones A & B (cleanup is only safe once real paths exist); coordinates with Issues 1, 3, 4, 12.

Size

S

Derived from the LLM narration gap analysis (Volo) and decomposition (Elminster).