Add tool_search support and compatibility alignment

[PenguinDOOM]

Summary

• add VS Code tool_search support for Responses API traffic
• introduce client compatibility mode handling across OpenAI, Ollama, and websocket routes
• align compatibility tests and retry/error handling across the affected surfaces

Compatibility Notes

• ChatMock normalizes Copilot/Codex Responses payloads for chatgpt.com/backend-api/codex/responses rather than trying to fully mirror the public OpenAI Responses API
• standard OpenAI-compatible tools support remains available in every mode, while VS Code-specific non-standard shims are gated behind client-compat=vscode
• truncation is accepted from clients but stripped before the upstream request because the Codex backend rejects that parameter
• store is always pinned to false because Copilot traffic is treated as non-persistent here and the upstream contract requires false

Added In This Update

• rebaselined the route suite around the current VS Code Responses tool_search contract, with native Responses coverage kept separate from legacy compatibility coverage
• added targeted regression coverage for native tool_search, tool_search_call, canonical tool_search_output, explicit previous_response_id passthrough, and deferred namespace follow-up on the HTTP responses route
• added a streaming regression test that locks HTTP SSE passthrough for response.output_item.done events carrying tool_search_call output
• confirmed the current production implementation already satisfies the required Responses contract, so this follow-up is test-focused rather than a production behavior change

Notes

• excludes the untracked .serena/ path

[Copilot]

Pull request overview

This PR adds VS Code-oriented tool_search support and introduces a CLIENT_COMPAT mode to gate/align behavior across the OpenAI, Ollama, and Responses surfaces, with expanded test coverage for these compatibility rules.

Changes:

• Add CLIENT_COMPAT configuration (default vs vscode) wired through create_app() and the CLI.
• Normalize/convert Chat Completions-style function tools (including tool_search) into Responses-style tool schema, and enforce compat gating for responses_* extensions.
• Extend route and model-registry tests to validate unknown model passthrough, upstream error passthrough, and compat-specific behavior (HTTP + websocket).

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
tests/test_routes.py	Adds extensive compatibility + tool conversion tests, including websocket coverage and server startup helper.
tests/test_models.py	Adds tests for unknown-model passthrough and model listing expectations.
chatmock/routes_openai.py	Adds compat gating for responses_* fields and improves upstream error JSON passthrough.
chatmock/routes_ollama.py	Adds compat gating for responses_* fields and refactors upstream response handling.
chatmock/responses_api.py	Adds CLIENT_COMPAT checks, tool schema detection, tool conversion, and strips additional unsupported fields.
chatmock/model_registry.py	Adds public_aliases support and includes them in public model listings/alias normalization.
chatmock/cli.py	Adds --client-compat flag and plumbs it into app creation.
chatmock/app.py	Adds CLIENT_COMPAT config to the app config defaults.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.