fix: use Responses API for DeepSeek by default

[smallmain]

Summary

• switch the built-in DeepSeek preset from OpenAI Chat Completions to OpenAI Responses
• keep the existing DeepSeek base URL, FIM beta endpoint, balance integration, and model list unchanged
• add a catalog regression test and update the English and Chinese manual configuration docs

Rationale

DeepSeek's current official documentation states that its API supports the Responses API at https://api.deepseek.com and documents the supported stateless request/streaming format:

https://api-docs.deepseek.com/guides/responses_api/

The project already has a compatible OpenAI Responses provider. The built-in DeepSeek preset was still selecting openai-chat-completion, so one-click configuration could not use the newly supported interface by default.

Validation

• npx vitest run test/unit/plan4-provider-catalog.test.ts -t 'uses Responses by default for DeepSeek' - passed
• npm run compile - passed
• npm run verify:chat-lib - passed
• npm run l10n:check - passed
• git diff --check - passed
• npm run test:unit - TypeScript and release checks passed; 103 test files and 1232 tests passed, with one pre-existing catalog assertion failing

The unchanged failing assertion expects every deepseek-* model to declare FIM. Current main added deepseek-v4-flash-vision-exp without a FIM template in commit 616a3bd, so that existing assertion now fails independently of this patch. This PR deliberately leaves that unrelated catalog issue untouched.

Fixes #291