[FEATURE] add dither, impulsive noise detection, regularized inversion and near/far-field merging - #113

Adds to pytta.functions four tools for measurements that the ITA-Toolbox has in its measurement application (ita_dither, ita_impulsiveNoiseDetection, ita_invert_spk_minphase_regularization and ita_add_nearfield_farfield_measurements):

1. dither(signal, bits=24, kind='tri', quiet=False, seed=None): dither noise added before quantization to bits bits, rectangular, triangular or Gaussian (Lipshitz, Wannamaker and Vanderkooy, "Quantization and dither: a theoretical survey", J. Audio Eng. Soc. 40(5), 1992). With quiet, silent samples stay silent.
2. impulsive_noise(ir, excitation, noise_start, threshold=20): the part of a measured impulse response after noise_start, convolved again with the excitation, rebuilds the background noise of the recording; L_max_rms = 20 lg(max / rms) of that noise above 20 dB flags an impulsive event. It returns the flag, the level and the noise.
3. invert_spectrum(signal, freqRange, minimum_phase=False): the regularized inverse conj(H) / (|H|² + ε) (Kirkeby, Rubak, Nelson and Farina, 1999), with ε negligible inside freqRange and half the largest |H|² outside it, crossfaded over half an octave, the rule that ImpulsiveResponse uses for its regularization. With minimum_phase the inverse gets the minimum phase of its magnitude.
4. merge_near_far(near, far, crossover, order=4, port=None, port_ratio=1): a loudspeaker response from a near-field measurement below the crossover and a far-field one above it (Keele, "Low-frequency loudspeaker assessment by nearfield sound-pressure measurement", J. Audio Eng. Soc. 22(3), 1974). The near-field response, with a port response added for a vented box, is aligned in time and level with the far-field one and crossfaded with power-complementary Butterworth weights that add up to 1.

These ITA functions belong to an application of the ITA-Toolbox whose license is "all rights reserved", so only their help text was read and the functions are written from the references.

The tests in tests/test_measurement_tools.py print the measured values:

1. With triangular dither the total error of an 8-bit quantizer on a sine of 1.5 steps has the power step²/4 (1.001 of it) and no mean on either half wave (9.1e-4 step), as the theory predicts.
2. With an excitation of flat magnitude, L_max_rms is 12.8 dB for Gaussian noise alone and 33.2 dB when a click falls in the recording.
3. The inverse of a band-pass filter is exact inside the range (6.6e-9, and 9.7e-9 with minimum phase), and the minimum phase inverse holds more energy at every instant from its start than the zero phase one with the same magnitude.
4. A near-field response 20 dB louder and 100 samples earlier than the far-field one is merged back into the far-field one (1.9e-8).

Checked with Python 3.12, numpy 1.26 and scipy 1.17. The 7 failures of development (3 in test_h5io.py, 4 in test_pyttasave.py) are the same with and without this change.

Co-Authored-By: Claude Fable 5.1 noreply@anthropic.com