feat: replace convert_units and pre_process with a unified transforms pipeline

[turban]

Problem

The current codebase has two separate, ad-hoc mechanisms for post-download data transformation before the GeoZarr is written:

• convert_units: degC / convert_units: mm — implemented in feat: apply unit conversion during zarr build #78 with a hardcoded _UNIT_CONVERSIONS lookup table
• pre_process: ['deaccumulate_era5'] — defined in YAML but not yet implemented

Neither supports multiple ordered transformations, and adding new conversions requires changing core code rather than configuration.

Proposed solution

Replace both fields with a single transforms list in the dataset YAML, using the same dotted-path callable pattern already used by ingestion.function:

```yaml

era5_land temperature

transforms:

• function: climate_api.transforms.convert_units

era5_land precipitation

transforms:

• function: climate_api.transforms.deaccumulate_era5
• function: climate_api.transforms.convert_units
  ```

Each callable has the signature (ds: xr.Dataset, dataset: dict[str, Any]) -> xr.Dataset and is resolved at runtime via importlib, exactly like download functions. convert_units reads the existing units/convert_units fields (kept for STAC metadata). Transforms from external packages (e.g. dhis2eo) are supported without any changes to core code.

Changes required

• Add src/climate_api/transforms/ module with at least convert_units (replacing _UNIT_CONVERSIONS) and a placeholder/implementation for deaccumulate_era5
• Update build_dataset_zarr() in downloader.py to run the transforms pipeline instead of calling _apply_unit_conversion() directly
• Update era5_land.yaml to use transforms: entries, removing pre_process and keeping convert_units/units for STAC metadata only
• Remove the hardcoded _UNIT_CONVERSIONS dict and _apply_unit_conversion() from downloader.py