feat(ai): add Requesty provider

[Thibaultjaigu]

Summary

• add Requesty to the desktop and mobile AI provider selectors (settings and onboarding), next to OpenRouter
• configure the OpenAI-compatible endpoint at https://router.requesty.ai/v1 (no /v1 auto-append, same as OpenRouter)
• reuse the existing model discovery, connection test, and chat runtime paths
• detect Requesty endpoints (including regional hosts such as router.eu.requesty.ai) when importing provider settings

Same footprint as the Atlas Cloud provider (#621): AIProviderType, PROVIDER_CONFIGS, detectProviderFromUrl, the four provider lists, and URL helper tests.

How to test

1. Get a key at https://app.requesty.ai/api-keys
2. Settings, AI, add an endpoint, pick Requesty, paste the key
3. Fetch models, then chat with a model id such as openai/gpt-4o-mini

Validation

• vitest run src/utils/api.test.ts: 19 passed
• core test run: 81 files, 599 tests passed
• core tsc --noEmit: one existing error in src/reader/justified-text.test.ts, unrelated to this diff
• biome check on the touched files: same diagnostics as on main, none on the added lines
• git diff --check
• live, through buildProviderModelsUrl, testAIEndpoint and createChatModelFromEndpoint: model list 200 (762 models), connection test OK, chat with openai/gpt-4o-mini returned the expected reply

One note: the connection test sends max_tokens: 4, and OpenAI models behind Requesty require at least 16, so testing the connection with an OpenAI model returns a 400 while chat works fine. Testing with another model (for example google/gemini-2.5-flash) passes. I left the shared test request unchanged.

No README, public docs, logo, dependency, or lockfile changes. The desktop and mobile apps were not built locally; the app edits are one line list entries.

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