feat: add OpenAI and Gemini LLM providers

[aditya-gupta-dev]

Adds OpenAI and Google Gemini as cloud moment-detection providers, alongside the existing DeepSeek / Claude / Ollama.

Part of #5. Starting with these two; Groq, OpenRouter, and xAI can follow the same pattern in later PRs.

Changes

Backend (src-tauri/)

• llm.rs: new detect_candidates_with_openai and detect_candidates_with_gemini, following the existing per-provider pattern (own response
  structs, shared compact_segments + parse_candidate_json helpers, same prompt).
• lib.rs: "openai" / "gemini" arms in generate_candidates (key + model resolved from arg → env → default), and has_openai_key /
  has_gemini_key in environment_status.
• models.rs: two new EnvironmentStatus fields.
• .env.example: OPENAI_API_KEY, GEMINI_API_KEY.

Frontend (src/main.tsx)

• OpenAI / Gemini in the LLM Engine dropdown, onboarding cloud form, and status bar.
• Per-provider API key + free-text model selection, persisted to localStorage.
• Key/model threaded through the generate_candidates calls; provider-aware warning labels.

Also: fixes an invalid "\:" string escape in build_drawtext_filters (lib.rs) that failed to compile (unknown character escape) — a
r"\:" raw string. Needed for the crate to build.

Providers / defaults

• OpenAI: POST /v1/chat/completions, Bearer auth, JSON mode. Default model gpt-4o-mini.
• Gemini: generateContent, x-goog-api-key, responseMimeType: application/json. Default model gemini-2.5-flash.

Testing

• cargo check — passes
• npm run build (tsc + vite) — passes
• Manual: Settings → pick OpenAI/Gemini → enter key + model → Import → Transcribe → Find Viral Moments returns candidates.

Notes

• Keys stored in localStorage (matches existing app behavior).
• Model selection is a free-text field (like the existing Ollama model input), so new models don't require a code change.