refactor

[Ha-Ree]

Summary

This PR cleans up the OasisDataManager library with ergonomic improvements, bug fixes, and expanded test coverage. Changes span the public API, internal reader logic, storage backends, and exception naming.

---

Public API improvements

Short import paths

• oasis_data_manager/__init__.py now exports PandasReader, DaskReader, PyarrowReader, LocalStorage, AwsS3Storage, AzureABFSStorage directly, allowing from oasis_data_manager import PandasReader instead of importing from the full backend path.
• oasis_data_manager/filestore/__init__.py now exports all three storage classes and the config helpers, enabling from oasis_data_manager.filestore import AwsS3Storage.
• oasis_data_manager/df_reader/__init__.py re-exports all reader classes.

Storage backend module renames

• filestore/backends/aws_s3.py → filestore/backends/aws.py (canonical path)
• filestore/backends/azure_abfs.py → filestore/backends/azure.py (canonical path)
• The old paths are retained as backward-compatible shims that emit a DeprecationWarning on import, so existing code continues to work.

Exception rename

• OasisException renamed to OasisDataManagerException to better reflect the library it belongs to and avoid confusion with the same name used elsewhere in the Oasis platform.
• OasisException is kept as a backward-compatible alias (same class object) so nothing breaks.
• MissingInputsException updated to subclass OasisDataManagerException.

---

Bug fixes

Dask RecursionError on parquet reads

• OasisReader._read() now sets has_read = True before calling read_parquet() / read_csv(), wrapped in a try/except that resets the flag on failure.
• Previously the flag was set after the read, causing Dask's read_parquet() to re-enter _read() via self.df (a property that calls _read()), producing infinite recursion.

Dask copy_with_df type mismatch

• OasisDaskReader.copy_with_df() now converts any incoming pandas DataFrame to a Dask DataFrame before passing it to the base implementation.
• Previously, copying with a pandas DataFrame left self._df as pandas, causing AttributeError when as_pandas() called .compute() on it.

Double _read() calls

• OasisReader.filter() and OasisReader.as_pandas() now access self._df directly instead of going through the self.df property, eliminating a redundant second _read() call.

---

Code quality

• Extension detection: Replaced a nested for/else/break loop in _read() with a one-line any() expression.
• Old-style super() calls: Updated to super() (no arguments) in AwsS3Storage, AzureABFSStorage, MissingInputsException, and OasisDataManagerException.
• f-strings: Replaced .format() call in MissingInputsException.__init__ with an f-string.
• Logger usage: delete_file() and delete_dir() in BaseStorage now call self.logger.info() instead of the bare module-level logging.info(), consistent with the rest of the class. Fixed a "Unknwon" typo in the log message.
• Stale docstring: Removed outdated Django-storage references and TODO stubs from AwsS3Storage.
• config_options serialization: AWS and Azure backends now store the original root_dir argument (self._root_dir_arg) before joining it onto the bucket/container path, and use it in config_options. This avoids a fragile Path.relative_to() reverse-computation that could fail if the paths didn't align.
• ComplexData.run() clarity: Added a comment explaining the fetch_required logic — CSV and Parquet files are read directly by the df_reader, so fetch() is only needed for formats the reader cannot handle directly.

---

Test coverage

New test files

• tests/df_reader/test_pyarrow.py: PyArrow backend tests covering parquet reads, column selection, and filter predicates.
• tests/filestorage/test_storage_utils.py: Tests for BaseStorage.create_traceback() and AwsS3Storage._strip_signing_parameters().

New tests in existing files

• test_read_csv.py / test_read_parquet.py: OasisReader.query(), copy_with_df(), and OasisDaskReader.read_from_dataframe().
• test_from_dataframe.py: Passing a pandas DataFrame via dataframe= to a Dask reader.
• test_caching.py: OasisDataManagerException backward-compat alias verification.

Test fixes

• Dask query() test: added .compute() call for lazy scalar results (e.g. frame["D"].sum()).
• test_complex/test_base.py: guarded dask import with pytest.importorskip so the file is skipped cleanly when Dask is not installed.
• mypy type: ignore comments on optional-dependency fallback assignments updated to suppress the correct error codes.

Readme creation

• Updated readme to be nonempty