Update yield Africa configs and add earthaccess dependency

[robknapen]

What does this PR do?

Update yield Africa configs and add earthaccess dependency

Configurations

• Model configs — standardize_targets: true enabled on all yield Africa regression models (geoclip, tabular, tessera, tessera_fusion). Targets are now z-scored before the loss; metrics are reported in original yield units. Comments on normalize_features updated to clarify it disables L2 unit-sphere projection, not feature scaling.
• yield_fusion_reg — GeoClip encoder branch now has an MLP projector (nn_layers=1, output_dim=256) to align its output dimension with the tabular branch before fusion.
• Metrics config — MAPE added to yield_africa_regression.yaml alongside the existing MAE, RRMSE, and R².
• Data configs — csv_name: null added explicitly to all data configs (all, loco, spatial, tessera, tessera_loco, tessera_spatial) to expose the new parameter from BaseDataset. Two split file defaults updated: spatial → split_spatial_50km.pth / split_spatial_100km.pth, LOCO default → split_loco_TAN.pth. use_country_features: false added to the LOCO config and year 2014 excluded.
• pyproject.toml — geotessera moved to its own optional extra (uv sync --extra geotessera). earthaccess and pyhdf added under create-data for MODIS data download. scipy-stubs and xgboost added as core dependencies. - .env.example — NASA EarthData credential placeholders and macOS pyhdf/HDF4 build instructions added.

Before submitting

• Did you make sure title is self-explanatory and the description concisely explains the PR?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you list all the breaking changes introduced by this pull request?
• Did you test your PR locally with pytest command?

[robknapen]

Depends on previous merge requests, so tests will fail until those are accepted.