Fix Spectrum slicing to preserve freq_axis metadata

[vishwas-droid]

This PR fixes an issue where slicing a Spectrum object could result in incorrect
or inconsistent propagation of its associated freq_axis metadata.

Since Spectrum subclasses numpy.ndarray, slicing operations return new views.
Without explicitly handling NumPy subclass mechanics, custom attributes such as
freq_axis may not always be reliably preserved.

Changes

• Implements __array_finalize__ to properly propagate metadata when new views are created.
• Overrides __getitem__ to ensure sliced Spectrum objects retain a correctly sliced freq_axis.
• Adds safe handling for advanced indexing (boolean masks, fancy indexing).
• Ensures consistent behavior between the data array and its corresponding frequency axis.

Behavior After This PR

• Standard slicing (spec[2:5])
• Boolean masking (spec[spec > 5])
• Fancy indexing (spec[[1, 3, 5]])
• Arithmetic operations (spec + 5)

All correctly preserve freq_axis.

Backward Compatibility

No breaking changes. Existing tests pass and behavior remains consistent for all supported use cases.