refactor: split out cache directories per provider and use consistent naming

[lewisjared]

Summary

Currently, all providers share a common cache directory at pooch.os_cache("climate_ref"). This can make it difficult to manage and debug provider-specific data. Additionally, data registries download to inconsistently named directories (e.g., pmp-climatology registry downloads to PMP_obs4MIPsClims/).

Current Behavior

• All providers use the same cache root: ~/.cache/climate_ref/
• Data registries download to directories determined by their internal structure, not their registry name
• Example: pmp-climatology registry → {cache_dir}/PMP_obs4MIPsClims/

Desired Behavior

1. Per-provider cache directories: Each provider should have its own cache subdirectory, e.g.:

   • ~/.cache/climate_ref/pmp/
   • ~/.cache/climate_ref/esmvaltool/
   • ~/.cache/climate_ref/ilamb/

2. Consistent registry naming: Data registries should download to folders matching their registry name:

   • pmp-climatology registry → {cache_dir}/pmp-climatology/
   • This makes it easier to correlate registry names with filesystem locations

Benefits

• Easier to debug provider-specific issues
• Clearer disk usage attribution per provider
• Simpler cleanup of individual provider data
• More predictable directory structure

Files to Update

• packages/climate-ref-core/src/climate_ref_core/dataset_registry.py - registry download paths
• packages/climate-ref-pmp/src/climate_ref_pmp/__init__.py - get_data_path() and ingest_data()
• Similar changes needed for other providers (esmvaltool, ilamb)

Notes

The ingest_data() method currently hardcodes PMP_obs4MIPsClims as the climatology subdirectory name. This should use the registry name (pmp-climatology) instead.