cabm - fix: Python 3.13 + updated dependency compatibility (pandas 2.2, PyTorch 2.11, gluonts 0.16, datasets 4.x)

[chandragupta]

Commit: Python 3.13 + updated dependency compatibility

Message: fix: Python 3.13 + updated dependency compatibility (pandas 2.2, PyTorch 2.11, gluonts 0.16, datasets 4.x)

Summary

Relaxed version pins in pyproject.toml to support Python 3.13 (numpy, scipy, torch, gluonts, datasets). Fixed resulting breakages across the codebase:

• src/uni2ts/transform/patch.py: Updated DEFAULT_RANGES keys to pandas 2.2+ frequency aliases (H→h, T→min, M→ME, etc.)
• src/uni2ts/data/builder/simple.py: Replaced Dataset.from_generator with Dataset.from_list to avoid Windows Arrow file-locking in datasets 4.x; added .astype("float32").squeeze() for stricter schema enforcement
• src/uni2ts/model/moirai2/forecast.py: Added _QuantilePredictionNetWrapper to adapt Moirai2Forecast.forward() output to the gluonts 0.16.x QuantileForecastGenerator interface (now expects ((tensor,), loc, scale))
• test/module/test_attention.py: Updated all-masked attention assertion for PyTorch 2.11 behaviour change
• test/data/builder/test_simple.py: Added explicit del hf_dataset before shutil.rmtree to release Arrow file handles on Windows
• test/transform/test_patch.py: Fixed frequency alias lookup to use norm_freq_str
• README.md: Fixed Getting Started code block (added missing moirai2 imports, removed unused imports, replaced hardcoded lengths with variables)
• test/sample_test_run.py: Added end-to-end smoke test; uses forecast.quantile("p50") instead of forecast.mean to avoid gluonts QuantileForecast warning

[salesforce-cla]

Thanks for the contribution! Before we can merge this, we need @chandragupta to sign the Salesforce Inc. Contributor License Agreement.

[chandragupta]

Thanks for the contribution! Before we can merge this, we need @chandragupta to sign the Salesforce Inc. Contributor License Agreement.

Already signed the CLA but the PR is wrongly labeled as cla:missing