feat: add plotting subpackage for radar dataset diagnostics

[franchg]

Summary

• Adds mlcast_datasets.plotting with 7 modules for radar dataset diagnostics:
  • domain_map — domain overview map with spatial coverage overlay
  • monthly_cycle — monthly precipitation climatology boxplot
  • precipitation_stats — mean/max/std maps + value histogram
  • sample_precipitation — precipitation event snapshot grid
  • spatial_coverage — data coverage fraction heatmap
  • temporal_coverage — monthly completeness heatmap + yearly bar chart
  • summary_table — metadata summary returning pd.DataFrame (CSV-saveable)
• Makes plotting an optional install: pip install 'mlcast-datasets[plotting]'
• Adds [tool.isort] profile = "black" to resolve pre-commit hook conflict

Test plan

• uv run pytest src/mlcast_datasets/tests/ -q — all tests pass
• uv pip install -e . — installs without plotting extras
• python -c "import mlcast_datasets" — core import works
• uv pip install -e ".[plotting]" — installs plotting extras
• python -c "from mlcast_datasets.plotting import plot_domain_map" — plotting import works

[franchg]

EXAMPLE OF USAGE

import mlcast_datasets
from mlcast_datasets.plotting import  plot_domain_map, plot_monthly_cycle, \
    plot_precipitation_stats, plot_sample_precipitation, plot_spatial_coverage, \
    plot_temporal_coverage, generate_summary_table

cat = mlcast_datasets.open_catalog()
ds = cat.precipitation.it_dpc_sri_5min.to_dask()
_ = plot_domain_map(ds.sel(time=slice("2025-01-01", None)), n_coverage_samples=1000)

_ = plot_spatial_coverage(ds, n_samples=10000)

_ = plot_monthly_cycle(ds, n_samples=10000)

_ = plot_sample_precipitation(ds, time_slice=slice("2023-07-01", "2023-07-02"), time_spacing_hours=1)

_ = plot_temporal_coverage(ds)

_ = plot_precipitation_stats(ds.sel(time=slice("2020-01-01", None)))

generate_summary_table(ds)

	Property	Value
0	Time range	2010-01-01 to 2025-12-31
1	Total timesteps	1,039,785
2	Missing timesteps	15,530 (1.5%)
3	Grid dimensions	1400 × 1200 pixels
4	Spatial resolution	1 km
5	Data variable	RR (Total precipitation rate)
6	Units	kg m-2 h-1
7	Data type	float32
8	Uncompressed volume	7.0 TB
9	Compressed volume	N/A
10	Compression ratio	N/A
11	Temporal frequency	15min (2010-01-01-2014-06-25), 10min (2014-06-...
12	CRS	Transverse Mercator
13	License	CC-BY-SA-4.0

[leifdenby]

Could you add docstrings throughout? I will give it a thorough review once I have those :)

[leifdenby]

This looks great @franchg :)

I have made a PR #37 which ensures that we build notebooks in CI in PRs (that doesn't happen now, I had overlooked that) and it also will comment with a link to this preview build. Maybe we could merge that first and then we can check with that how long your notebooks take to build?

[github-actions]

View preview of built jupyterbooks on https://mlcast-community.github.io/mlcast-datasets/pr-preview/pr-30/
(preview is automatically rebuilt and uploaded on later commits)

[leifdenby]

The rendered notebooks look great @franchg, but it takes the execution of the jupyterbook build from ~ 4min (https://github.com/mlcast-community/mlcast-datasets/actions/runs/24443479747/job/71413893785) to ~15min (https://github.com/mlcast-community/mlcast-datasets/actions/runs/24993116111/job/73183302255)

Maybe we need to think about how to can reduce the long-running computations a bit? Otherwise we need to work out how to execute the notebook build closer to the data (i.e. on a EWC host)