Align ntfperf integration grid

[chenzc24]

Summary

• Add explicit NTF integration grid controls to ntf_analyzer.
• Keep ntf_analyzer's historical Python grid as the default.
• Make ntfperf use MATLAB ntfperf.m's grid contract by default: w = (1:N)/N/2, N = 1_000_000.
• Update oversampling compatibility tests to check the explicit MATLAB grid policy instead of treating ntfperf as a simple alias.

Rationale

This is a grid-policy alignment, not a core algorithm change. Python and MATLAB already use the same integrated squared-magnitude formula:

np = sum(|NTF(w)|^2 in band) / len(w)
snr = -10*log10(np)

The observed mismatch comes from the integration grid:

MATLAB ntfperf: w = (1:N)/N/2, N = 1e6
Python ntf_analyzer: np.linspace(0, 0.5, 2**16)

ntf_analyzer remains the lightweight Python analyzer. ntfperf is the MATLAB-compatible wrapper, so it now defaults to MATLAB's grid while still exposing n_grid for tests and experimentation.

Experiment

Dataset/function cases generated from MATLAB ntfperf.m and Python on the same NTFs:

case	Python ntf_analyzer default	delta vs MATLAB	Python ntfperf MATLAB grid	delta vs MATLAB
[1,-1]/[1,0], band [0, 0.5/16]	30.9616429351113	+0.001352172349	30.9602907627623	+2.84e-14
[1,-2,1]/[1,0,0], band [0, 0.5/16]	47.3320195872251	+0.002208366542	47.3298112206832	-2.84e-14
[1,0,1]/[1,0,0], band [0.24, 0.26]	36.7669311709645	-0.003238854921	36.7701700258856	0

Tests

• uv run --with pytest pytest tests/unit/oversampling/test_matlab_compat_api.py -q
• uv run --with pytest pytest tests/unit/oversampling/test_noise_shaping_examples.py -q
• uv run --with pytest pytest tests/unit/oversampling/test_ntf_analyzer.py -q -m "not slow"
• uv run --with pytest pytest tests/unit/oversampling/test_matlab_compat_api.py tests/unit/oversampling/test_noise_shaping_examples.py tests/unit/oversampling/test_ntf_analyzer.py -q -m "not slow"
• git diff --check