feat: streaming CSV summarization + environment diagnostics and desktop UI integration

[rad1092]

Motivation

• Improve memory usage and support large CSVs by switching summarization to a streaming-style collector.
• Provide a quick local environment check for Ollama and model availability to make diagnoses easier before running models.
• Expose diagnostics from both the CLI and the Windows desktop UI so users can validate runtime dependencies without leaving the app.

Description

• Implemented streaming summarization by adding summarize_reader and routing build_analysis_payload/build_analysis_payload_from_csv_text to use it in bitnet_tools/analysis.py.
• Added a new diagnostics module bitnet_tools/doctor.py with collect_environment(model) to check Python/platform info, ollama in PATH, ollama --version, and ollama list + model availability.
• Exposed diagnostics via a new CLI subcommand doctor in bitnet_tools/cli.py and added a desktop command path and bitnet-desktop entry point in pyproject.toml.
• Integrated a GUI 환경진단 button into the Windows desktop app (bitnet_tools/desktop.py) to show diagnostics in-app and added a thin bitnet_desktop.pyw launcher plus BitNet_Desktop_Start.bat for one-click start on Windows.
• Updated README.md to reflect the diagnostics flow, Windows start guidance, and updated completion status.
• Added/updated tests for streaming mixed-type handling and the doctor CLI output in tests/test_analysis.py and tests/test_cli.py.

Testing

• Ran pytest -q and all tests passed (7 passed).
• Executed the diagnostic CLI with python -m bitnet_tools.cli doctor --model bitnet:latest, which ran successfully and produced a valid diagnostics report (in this environment it reported ollama not found in PATH).

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

First_two/bitnet_tools/desktop.py

Line 130 in df2042a

question = self.question.get("1.0", "end").strip()

Keep Tkinter widget access on the main thread

_analyze_async, _doctor_async, and _run_model_async run work in background threads, but those workers call Tk widget APIs (self.question.get() here, and similarly self.model.get() in other handlers). Tkinter is not thread-safe, so this can intermittently raise Tcl/thread errors or hang the desktop app when users click Analyze/Doctor/Run. Capture widget values on the UI thread before starting the worker, and only do non-Tk work inside the thread.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".