Data layer: normalisation stats, custom CSV name, LOCO country features

[robknapen]

What does this PR do?

Data layer: normalisation stats, custom CSV name, LOCO country features, for CY use case

• Normalisation stats in BaseDataModule: After splitting, the datamodule now computes per-feature and per-target mean/std on the training fold (_compute_tabular_normalisation_stats, _compute_target_normalisation_stats). Stats are exposed as abular_normalisation_stats and target_normalisation_stats for use by encoders/heads. Constant features are clamped to std=1 to avoid division by zero.
• Custom CSV filename in BaseDataset: Added csv_name parameter so a dataset can load from a non-default CSV (e.g. a country-specific file) without subclassing. Falls back to model_ready_<dataset_name>.csv when not set.
• LOCO-safe country features in YieldAfricaDataset: Added use_country_features flag (default True). When False, country one-hot columns are not injected. This is needed for leave-one-country-out evaluation where the held-out country's one-hot is always 0 in training and always 1 at test time — a guaranteed distribution shift that inflates error.
• Vectorised file-path assignment: Replaced a slow row-by-row loop with a pd.concat + apply approach in BaseDataset._set_eo_paths.
• Test fixtures now respect the --use-mock CLI flag instead of hardcoding mock=True.
• Added test_yield_africa_dataset_no_country_features to verify no feat_country_* columns appear when the flag is off.

Changes are all CY dataset specific, and coded as opt-ins, so should not break existing code.

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?

[gabrieletijunaityte]

These were great insights on some bugs!