[Arcadia-1]

Summary

compute_spectrum / plotspec
  find the strongest visible spur peak
  if the full spur lobe is visible
    integrate it directly
  if signal/DC/band-edge masking hides part of the lobe
    restore the missing power from the measured fundamental lobe/peak ratio

plotspec sideBin='auto'
  detected width = max(noise-floor crossing, window coherent-main-lobe width)

This keeps the integrated-lobe SFDR convention established by #64/#65/#67, removes the 0.79–1.76 dB optimistic boundary bias, and prevents a noise-free Hann record from collapsing to sideBin=0.

Fixes #100.

Evidence

• Before: issue SFDR: spur power truncated by masked bins (optimistic up to ~1.8 dB) and definition differs from MATLAB plotspec #100's six-case Python repro reported 80.79 dBc at one masked neighbor and 81.76 dBc when both neighbors were unavailable.
  After: all six cases report 80.00 dBc, including public defaults with automatic side_bin.
• Added Python regressions for signal-mask, DC, Nyquist, double-edge, full-lobe control, and assumed_sig_pwr_dbfs behavior.
• Added matching MATLAB regression coverage for boundary compensation and noise-free Hann auto-sideBin behavior.
• Python validation: 901 passed, 11 deselected, 1 xfailed; integration validation: 34 passed.
• MATLAB/Octave is not installed on the development host, so the MATLAB regression is committed but was not executed locally.

Merge Danger

Door: Two-way

The change is isolated and can be reverted without data migration or API rollback.

Blast Radius: Spectrum metrics

Only SFDR cases whose spur lobe touches the signal/DC/analysis boundary change; MATLAB automatic side-bin selection also changes for noise-free windowed tones.