feat(weather): add precipitation and climate diagnostics

[stacknil]

Summary

• Extend the weather diagnostics toolkit with public-safe methods abstracted from the local source-document themes.
• Add station missing-value handling, IDW station-to-grid interpolation, accumulated-precipitation conversion, event totals, and threshold masks.
• Add climate-statistics helpers for anomalies, standardized anomalies, composites, and grid-point Pearson correlations.
• Add moisture flux divergence, specific-humidity alias support, two synthetic CLI summaries, expanded examples, and reviewer-facing documentation.

Validation

• python -m ruff check .
• python -m pytest -> 30 passed
• python -m compileall src scripts
• CLI help smoke checks for thermodynamic, dynamics, precipitation, climate-statistics, and ensemble scripts
• Synthetic precipitation and climate-statistics scripts executed successfully

Privacy and Data Boundary

• Candidate-file scan returned 0 restricted artifact files, 0 text privacy matches, and 0 non-ASCII text files.
• No raw weather datasets, station files, course documents, compressed archives, generated binary artifacts, original report text, or local-machine materials are included.
• New examples are synthetic and documented as synthetic.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 290e6a4f8a

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]

Use exact sentinel matching for missing-value code

mark_missing_sentinel uses np.isclose(array, sentinel) with NumPy’s default relative tolerance, which is very wide for a large sentinel like -32766 (about ±0.32766). That means valid values near the sentinel (e.g., -32765.8) are silently converted to NaN, corrupting downstream interpolation/threshold diagnostics whenever source data are scaled or rounded around the missing code. For sentinel-coded data this should be exact equality (or at most a tiny absolute tolerance) rather than relative tolerance.

Useful? React with 👍 / 👎.