Fix KeyError when accessing metadata for PSP RFS CDF files

[MohamedAli1937]

PR Description

Fixes #184

Summary:
This PR resolves a key mismatch between SpectrogramFactory and the RFSSpectrogram class that causes a KeyError when attempting to access metadata properties on spectrograms loaded from real Parker Solar Probe (PSP RFS) CDF files.

Motivation and Context:
Currently, SpectrogramFactory reads global attributes from CDF files using cdflib and passes them into the final metadata dictionary under the key "cdf_globals".
However, properties in RFSSpectrogram (specifically level and version in psp_rfs.py) attempt to read that data using self.meta["cdf_meta"].

Because of this mismatch, downloading a real PSP RFS file via Fido and calling spec.level throws an instant runtime KeyError: 'cdf_meta'. (The existing unit tests pass only because they mock the parse_path return value and manually inject mock metadata using the "cdf_meta" key.)

Changes:

• Modified the output dictionary key from "cdf_globals" to "cdf_meta" in spectrogram_factory.py (for both PSP and SOLO RPW parsing blocks) to align the factory output with the keys expected by the source classes and the existing test suite.

Architectural Justification: Why modify the Factory instead of psp_rfs.py?
It might intuitively seem safer to change psp_rfs.py to look for "cdf_globals". However, surgically modifying spectrogram_factory.py is the strictly correct choice for the following reasons:

1. Test Suite Compatibility: The existing test suite (e.g., test_psp_rfs.py, and test_solo_rpw.py) explicitly mocks metadata using the "cdf_meta" key.
2. Zero-Regression Risk: A codebase-wide search confirms that the string key "cdf_globals" was used exclusively inside the factory output.
3. Separation of Concerns: cdf_globals represents the raw output from the cdflib.globalattsget() function. Renaming the final exported dictionary key to "cdf_meta" abstracts the data layer away from low-level cdflib terminology, which is a cleaner architectural design.

Note on Implementation: The local Python variables inside the factory function remain named cdf_globals. This is intentional: they correctly describe the local return of cdflib, and leaving them as-is minimizes unnecessary line changes (diff noise) in the PR. Only the exported dictionary key string was updated (i.e., "cdf_meta": cdf_globals).

AI Assistance Disclosure

AI tools were used for:

• Code generation (e.g., when writing an implementation or fixing a bug)
• Test/benchmark generation
• Documentation (including examples)
• Research and understanding
• No AI tools were used

Regardless of AI use, the human contributor remains fully responsible for correctness, design choices, licensing compatibility, and long-term maintainability.

[samaloney]

Looks good but could you add a failing test using real data which this PR fixes?

[MohamedAli1937]

I added a real data test (test_psp_rfs_real_data) using @pytest.mark.remote_data and Fido.fetch.

While doing this, I found a hidden bug: mock data used "Data_type" as a string, but real CDF data (via cdflib) returns it as a list (e.g., ["L2>Level 2 Data"]). This caused: AttributeError: 'list' object has no attribute 'split' in psp_rfs.py.

Fixes in this PR:

• Added the real data test to validate spec.level and spec.version.
• Updated RFSSpectrogram to handle list values safely.

All tests pass locally, including the new real data test.

To run the test:

pytest --remote-data=any radiospectra/spectrogram/sources/tests/test_psp_rfs.py

Please have a look @samaloney, thanks!