Ensure freq_axis metadata propagation during NumPy ufunc evaluation

[vishwas-droid]

Spectrum objects were losing their freq_axis metadata when passed through NumPy ufuncs
(e.g., np.sqrt, np.log, arithmetic operations such as spec + 2).

Although the resulting object remained a Spectrum instance, the associated metadata
was silently dropped, which could lead to incorrect downstream scientific interpretation.

What This PR Does

This PR implements a custom __array_ufunc__ override to:

• Execute NumPy ufuncs on plain ndarrays
• Wrap the result back into a Spectrum
• Safely propagate freq_axis metadata (deep copy)
• Avoid recursion and ndarray subclass pitfalls

Tests

Regression tests have been added to ensure:

• freq_axis is preserved after unary ufuncs (e.g., np.sqrt)
• freq_axis is preserved after binary operations (e.g., spec + 5)

These tests prevent future silent metadata loss.

Rationale

Preserving metadata integrity is critical in scientific workflows, especially
for radio spectrogram analysis where frequency alignment is essential.