feat: strengthen CSV analysis safety, error handling, and robustness

[rad1092]

Motivation

• Make CSV ingestion more robust against empty input, missing headers, mixed types, different encodings and large UI payloads.
• Prevent long-running external calls by adding timeouts to ollama invocations from CLI and web UI.
• Provide richer numeric diagnostics (IQR-based outlier counts) to improve downstream prompt clarity.

Description

• Added AnalysisError and explicit validation in bitnet_tools/analysis.py to surface errors for empty CSV content and missing headers, and trimmed/validated header names.
• Implemented encoding fallbacks when reading files (utf-8-sig, utf-8, cp949) and unified CSV text parsing with csv.Sniffer delimiter detection.
• Strengthened dtype inference (int, float, date, string) and expanded numeric stats to include outlier_count (IQR rule) alongside count/mean/min/q1/median/q3/max/std in summarize_rows.
• Exposed AnalysisError and core helpers in bitnet_tools/__init__.py and added CLI (bitnet_tools/cli.py) with --timeout for ollama and friendly error handling for analysis failures.
• Added a simple local UI server (bitnet_tools/web.py) with payload size guard (MAX_CSV_TEXT_CHARS) and ollama timeout support, plus static UI assets (ui/*.html/.js/.css).
• Added pyproject.toml and expanded test coverage and test file (tests/test_analysis.py) to cover semicolon-delimited CSVs, empty input, mixed-type column behavior, and presence of outlier_count.

Testing

• Ran unit tests with python -m pytest -q which returned 5 passed.
• Performed a CLI smoke test with a semicolon-delimited CSV using python -m bitnet_tools.cli analyze /tmp/next2.csv --question '요약' --out /tmp/next2_payload.json and verified inferred dtypes were {'id': 'int', 'value': 'int', 'dt': 'date'} and outlier_count printed as 0.0.
• Verified web API safeguards by adding a maximum CSV text size and that AnalysisError is returned as a clear HTTP error when input is invalid.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 91ecdaea92

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".

[chatgpt-codex-connector]

Normalize row keys after stripping CSV headers

_parse_csv_text trims header names into columns but returns raw DictReader rows unchanged, so padded headers like " amount " are later queried as row.get("amount") and treated as entirely missing in summarize_rows. This silently corrupts dtype/missing/stat outputs for otherwise valid CSV files that contain whitespace around header names; map row keys to the normalized header names before returning.

Useful? React with 👍 / 👎.