[FEATURE] add time_window for SignalObj - #49

Adds pytta.time_window(signalObj, interval, unit='time', window='hann', symmetric=False).

The window is built from pairs of bounds. start <= end is a fade out (ones, falling half window, zeros); start > end is a fade in; several pairs multiply, so the usual impulse response window is [fadeInEnd, 0, fadeOutStart, fadeOutEnd]. symmetric=True mirrors the window about the centre of the signal, for slightly non causal impulse responses. Bounds are given in seconds or in zero based samples. The slope has abs(end - start) + 1 samples and is one half of a symmetric window twice as long, so a hann slope ends at exactly zero. The input is not modified; a new SignalObj with the same metadata is returned.

The behaviour follows ita_time_window of the ITA-Toolbox (RWTH Aachen), except that the bounds are not moved to even sample numbers. The implementation is written from the documented behaviour; no ITA code is copied.

Tests (tests/test_time_window.py, 9 tests):

• shape of the falling slope against the analytic symmetric Hann formula
• fade in as the mirror image of the fade out
• composition of several pairs by product
• rectangular window as a hard cut
• seconds to samples conversion
• symmetric option
• new object, metadata kept, every channel windowed the same way
• rejection of invalid interval, unit and window
• sample by sample comparison with the ITA-Toolbox on five cases (fade out, fade in, both ends with hann and with rectwin, symmetric). Reference data: tests/golden/time_window_ita.mat (79 kB), generated by tests/golden/make_time_window_ita.m with MATLAB and the ITA-Toolbox. Measured difference: 5e-8 of the signal peak, the float32 floor of SignalObj.

Checked with Python 3.12, numpy 1.26, scipy 1.17. The existing tests behave as on development: the 7 failures there (3 in test_h5io.py, 4 in test_pyttasave.py) are the same with and without this change. NumPy 2 support is in #53.

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