fix: make Checker and Hopper compatible with tool-based structured output

[lihujun101]

Checker reports and Hopper package-name extraction can fail with HTTP 400 when an OpenAI-compatible endpoint rejects the default response_format: json_schema. Use function_calling with tool_choice="auto" for Checker and both Hopper model paths.

Checker now explicitly requests the CheckReport tool and completes all tool-result messages before appending user-carried screenshots, preserving tool-call/result ordering.

Scope: only artemis/agents/checker/checker.py and artemis/agents/hopper/hopper.py; no other fixes from the fork's fix-bugs branch are included.

Validation:

• 46 targeted tests passed (Checker, structured reports, launch-app tools, and structured LLM handling).
• Local HTTP transport checks confirmed that both Hopper paths omit response_format and parse tool output. No live provider validation was performed.
• Formatting, lint for both changed files, quality ratchet, and typecheck passed.
• Full deterministic suite: 2,107 passed, 82 failed, 6 skipped, 8 deselected in the local environment. Failures include mocked provider configuration, summarizer/video tests, and MCP platform/schema tests; the full suite is not green.
• Repository-wide lint reports 11 findings in unchanged playground/backend_manager files.