feat(ai-providers): add Requesty as an AI provider

[Thibaultjaigu]

Adds Requesty, an OpenAI-compatible LLM router, as an AI provider. It is wired exactly like OpenRouter and goes through the same OpenAICompatibleProvider transport, so there is no new client code.

Changes

• AIModels.swift: new requesty case on AIProviderType, placed after openRouter, with display name, default endpoint https://router.requesty.ai (resolves to /v1/chat/completions and /v1/models) and the arrow.triangle.branch symbol. Auth and endpoint style fall through to the defaults OpenRouter uses (API key, chat completions).
• AIProviderRegistration.swift: .requesty added to the OpenAI-compatible family loop next to .openRouter, so it gets the same capabilities (chat, models, reasoning, images, configurable Base URL, max output tokens, model list fetch).
• Tests: .requesty added to the HTTP API-key provider capability test, and an endpoint resolution case next to the OpenRouter one.
• docs/security/privacy.mdx: router.requesty.ai added to the list of preset endpoints.
• CHANGELOG.md: entry under Unreleased, Added.

It is not the default provider and is only used when picked from Add Provider.

How to test

1. Settings > AI > Add Provider > Requesty.
2. Paste a key from https://app.requesty.ai/api-keys (the key is sent as Authorization: Bearer).
3. Pick a model from the fetched list or type one, for example openai/gpt-4o-mini or anthropic/claude-sonnet-4-5, then Test Connection.
4. EU routing: set the Base URL to https://router.eu.requesty.ai.

Checks

• swiftlint lint --strict on the four touched Swift files: clean.
• docs/scripts checks (writing style, docs against source, links): all pass.
• Not built locally: this machine has no Xcode, only the Command Line Tools with Swift 5.10, so I could not run xcodebuild build or the test target. I did a manual compile review instead: the three exhaustive switches over AIProviderType (displayName, defaultEndpoint, symbolName) each get a .requesty arm, authStyle and endpointStyle keep their default: arms, and no other switch names openRouter. AIProviderCapabilitiesTests.everyTypeHasDescriptor and AIEndpointTests.resolvesEveryDefaultEndpoint cover the new case through allCases.
• Live check: I compiled OpenAICompatibleProvider.swift unchanged together with AIModels.swift, AIEndpoint.swift and the chat wire types it depends on into a small command line harness, built an AIProviderConfig(type: .requesty, model: "openai/gpt-4o-mini") and used its endpoint. fetchAvailableModels() returned 762 models, testConnection() returned true, and streamChat streamed a reply with usage (14 input, 4 output tokens).

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