Attribute hygiene in S2 output: flag_meanings, short_name, dtype, valid_min/valid_max #351

Problems

1. flag_meanings is a JSON list. CF defines it as one string of words separated by blanks. 36 arrays in one product.
   • We write the SCL list in SCL_ATTRS (s2_optimization/s2_multiscale.py:51-54).
   • We pass through the source lists on quality/mask/*, conditions/mask/l1c_classification/*/b00 and conditions/mask/detector_footprint/*.
2. short_name is wrong on other levels. Example: /measurements/reflectance/r60m/b08 has b08_r10m, and scl at r120m/r360m/r720m has scl_r60m.
   • inject_missing_bands copies the source attributes (s2_multiscale.py:280).
   • _coarsen_variable copies them again for each level (s2_multiscale.py:208).
   • The style is also mixed (b05_60m and b08_r60m).
3. dtype attribute. It repeats the Zarr data type. sanitize_array_attrs removes it only for decoded floats (conversion/utils.py:1158-1160). 108 arrays.
4. valid_min/valid_max. The docstring of sanitize_array_attrs (utils.py:1142-1145) says that they are removed for decoded floats, but the code does not remove them. The source values pass through: for example, scl is uint8 with valid_max: 65535.

Proposal

• In sanitize_array_attrs, join a list flag_meanings with blanks. Write SCL_ATTRS as a string.
• Write short_name again for each level, as <variable>_r<resolution>m. If no reader needs it, remove it.
• Always remove the dtype attribute.
• Remove valid_min/valid_max when the value is outside the range of the data type. Do what the docstring says for decoded floats.
• Do not change _FillValue (see #TBD-01).

Tests

• Regenerate tests/_test_data/optimized_geozarr_examples/*.json (REGENERATE_SNAPSHOTS=1).
• Add checks in tests/test_array_attrs.py:
  • flag_meanings is a string;
  • there is no dtype attribute;
  • valid_* is inside the dtype range;
  • short_name matches the level.

Impact

No reader found for these attributes. titiler /info shows a different minmax for SCL.