Retire legacy provider fallback modules — LiteLLM sole route

[esafwan]

Finding: D-07 (CodeDiscovery grill-with-docs session) · Base branch: feature/inline-video-playback @ eebb9dc (based on feature/huf-design-system; repo default: develop)

Summary

Runs route LiteLLM-first but keep a legacy per-provider fallback path. The legacy modules have no streaming and no cost reporting, and openai.py is a 4-line stub. Two execution paths for the same job.

Evidence

• run.py:16-78 — fallback branch.
• providers/anthropic.py, providers/google.py, providers/openrouter.py, providers/openai.py (4-line stub).
• discovery/03 §6 — legacy path lacks streaming/cost.

Context (owner decision)

Q14 — owner: "litellm should be only route, legacy retires". LiteLLM is the sole canonical route from a Run to a provider.

Proposed fix / acceptance criteria

• Confirm LiteLLM coverage for every provider currently configured (incl. any edge models the legacy path served).
• Remove the fallback branch in run.py and delete the legacy provider modules.
• All runs route via LiteLLM; streaming and cost reporting uniform across providers.