feat: collect online tools and references for offline-ready workflow

[rad1092]

Motivation

• Capture pip wheels and authoritative UI/accessibility references during an internet-enabled turn so offline hosts can later install required tooling and consult preserved docs.
• Centralize the list of downloadable packages and reference URLs to make collection reproducible and scriptable.
• Make numeric parsing and baseline fixtures stable for regression checks when normalizing human-entered numeric formats.

Description

• Add resources/online_sources.json to declare tool_packages and reference_urls for automated collection.
• Add scripts/collect_online_assets.py which downloads pip wheels to .online_assets/wheels, fetches reference HTML to .online_assets/references, and emits a structured report at .online_assets/meta/collection_report.json.
• Add ONLINE_RESOURCE_COLLECTION.md documenting the collector and update .gitignore to exclude .online_assets/ artifacts.
• Add scripts/prepare_online_bundle.sh to build a local project wheel, attempt optional dependency downloads, archive an Ollama installer script if reachable, and write an offline usage guide under .offline_bundle/.
• Add CSV fixtures under tests/fixtures/ and extend tests/test_analysis.py with test_to_float_normalizes_currency_comma_percent_parentheses and test_baseline_fixture_summaries_are_stable to assert numeric normalization and baseline behavior.
• Normalize _to_float in bitnet_tools/analysis.py to strip common currency symbols, thousands separators, percent signs, and treat values in parentheses as negative.

Testing

• Ran pytest -q and all tests passed (26 passed).
• Executed python scripts/collect_online_assets.py, which saved .online_assets/meta/collection_report.json showing successful downloads for configured pip packages and captured reference fetch failures (environment tunnel/proxy returned 403) in the report.
• Ran python -m bitnet_tools.cli doctor to validate CLI diagnostics and confirm environment status (reported ollama not found).

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

First_two/bitnet_tools/analysis.py

Line 42 in 329455d

v.replace(",", "")

Parse comma-decimal values without magnitude corruption

Unconditionally stripping commas causes locale-formatted decimals to be misread, so values like "12,5" become 125.0 instead of 12.5. Because _to_float feeds summarize_reader, this silently changes dtype detection and numeric stats (mean/min/max) rather than failing fast, which can corrupt analysis results for CSVs that use comma decimal separators.

---

First_two/scripts/prepare_online_bundle.sh

Line 39 in 329455d

if python -m pip download -r "${META_DIR}/requirements_online.txt" -d "${WHEEL_DIR}"; then

Enforce wheel-only downloads for offline bundle deps

This command claims to collect optional dependency wheels, but pip download without --only-binary=:all: can legally fetch source archives, which breaks the offline install flow on hosts without a build toolchain. python -m pip download --help documents --only-binary as “Do not use source packages,” so leaving it out here means the bundle is not reliably wheel-only.

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".