Normalize remote dataset file types from URLs

[biefan]

Summary

• normalize remote dataset file types before handler lookup
• ignore URL query strings and fragments when inferring the file extension
• add regression coverage for signed-style URLs and uppercase extensions

Problem

_RemoteDatasetLoader._fetch_from_url() currently infers the file type with source.split(".")[-1]. That breaks two real input shapes:

• public URLs with query strings, e.g. https://example.com/data.json?download=1
• URLs or paths with uppercase extensions, e.g. https://example.com/data.JSON

Both are currently rejected with Invalid file_type before the loader even reaches the HTTP fetch path, even though the underlying data format is valid and supported.

Testing

• .venv/bin/pytest tests/unit/datasets/test_remote_dataset_loader.py -q

[romanlutz]

This needs fixes to pre-commit steps