[mrhy7176]

Summary

Add an optional CloudGPU memory provider preset using the existing OpenAI Chat Completions transport, following the merged Runcrate preset in #457 (and #420 / #432 before it). CloudGPU serves models such as DeepSeek, GLM and Kimi over an OpenAI-compatible API.

• memoryProvider: "cloudgpu": endpoint default https://cloudgpu.app/v1, model default deepseek-v4.1-flash, CLOUDGPU_API_KEY fallback, explicit env:// key configuration, and a distinct cloudgpu session tag.
• Factory, configuration validation, readiness detection, example/template, and all README languages updated.
• Tests cover authentication, request serialization (tool_choice: "required"), tool response extraction, overrides, environment-key resolution, and rejection of remote-provider changes from project configuration.
• No defaults change for existing users. Users opt in with their own key (created at https://cloudgpu.app/api/console; API docs at https://cloudgpu.app/api/docs); the docs name the destination for conversation data (https://cloudgpu.app).

Files changed

• src/services/ai/providers/cloudgpu.ts (new): CloudGPUProvider extends OpenAIChatCompletionProvider, overriding endpoint, default model and session tag
• src/services/ai/ai-provider-factory.ts, src/services/ai/provider-config.ts, src/services/ai/session/session-types.ts, src/types/index.ts: wiring and types
• src/config.ts: provider union, CLOUDGPU_API_KEY fallback, readiness check, config template
• opencode-mem.example.jsonc, README.md, .github/i18n/README.{ar,de,nl,tr,zh}.md: docs
• tests/cloudgpu-provider.test.ts (new), tests/ai-provider-config.test.ts, tests/config-resolution.test.ts: tests mirroring the Runcrate ones

Test plan

• Exercised the change locally: not done. This PR was prepared through the GitHub API without a local checkout, so the new tests were written (mirroring the Runcrate tests from feat(ai): add Runcrate memory provider preset #457) but have not been run locally. The default model ID deepseek-v4.1-flash was confirmed against the public catalogue at GET https://cloudgpu.app/v1/models.
• Known limitations: same as other OpenAI-compatible presets; "forceToolChoice": false is documented for models that reject forced tool calls.

Checklist

• Branch is based on main (do not open PRs from commits already on main)
• bun test passes (not run locally; relying on CI)
• bun run typecheck passes (not run locally; relying on CI)
• bun run check passes (not run locally; relying on CI)
• Docs / README updated when behavior or contribution process changes

Disclosure: I work on CloudGPU; this PR was prepared with AI assistance.