feat(ai): add Requesty as an OpenAI-compatible provider

[Thibaultjaigu]

Summary

Adds Requesty as a BYOK AI provider. It is an OpenAI-compatible LLM gateway (base URL https://router.requesty.ai/v1), so it reuses the existing callOpenAICompatible path. Every place OpenRouter is registered got a Requesty sibling; no shared client code changed.

Changes

• src/store/types.ts, src/store/aiSettings.ts, src/store/aiSettingsSchemas.ts: requesty added to the AIProvider union, AI_PROVIDERS and the zod enum
• src/config/aiProviders.ts: base URL, default model (openai/gpt-4o-mini), provider meta (key link https://app.requesty.ai/api-keys), model list, default base URL, risk browser_friendly (the router answers CORS preflight with access-control-allow-origin: *)
• src/services/aiService.ts: VITE_REQUESTY_API_KEY env fallback
• src/components/SettingsModal/ai/AdvancedEndpointSection.tsx: same POST /chat/completions hint OpenRouter shows
• public/logos/requesty.svg: monochrome logo in the same format as openrouter.svg (works with the mask-image icon)
• _headers, nginx/nginx.conf: https://*.requesty.ai added to connect-src next to https://openrouter.ai
• src/i18n/locales/en and public/locales/en: English provider hint and model labels (other locales fall back to English), kept in sync
• README provider table and counts, docs-site AI generation page (en, tr)

Requesty is not the default provider and is not added to any automatic selection.

Model list: openai/gpt-4o-mini plus four Requesty managed policies (gpt-5.4-mini, claude-sonnet-4-6, gemini-3.5-flash, deepseek-v4-flash). Managed policy ids are short names that route across several upstream providers with failover. All ids checked against the live GET /v1/models and GET /v1/models/managed.

How to test

1. Set VITE_REQUESTY_API_KEY=<your key> in .env.local, or pick Requesty in Settings, AI and paste a key
2. Keep the default model openai/gpt-4o-mini (or pick another from the list)
3. Generate a diagram in Flowpilot

Validation

• npx tsc -b: pass
• npx eslint on the touched TS files with --max-warnings 0: pass
• npx vitest --run: 296 files, 1465 tests passed (includes the provider config consistency and locale coverage/sync tests)
• Live check through aiService (throwaway vitest file, not committed): chatWithFlowpilot with provider requesty streamed a reply from openai/gpt-4o-mini, and generateDiagramFromChat returned DSL for all five listed models

Disclosure: I work at Requesty. Happy to adjust anything to match project conventions.