emikit: ground-bounce CM estimator + Montrose envelope + cable CLI

[UnsignedChad]

Follow-up to #42. Three things landed together because they all had to be in for the TI re-validation to come out right.

1. Ground-bounce CM current estimator

CableSpec gains ground_inductance_h and cable_cm_inductance_per_m_h. estimate_cm_current(cable, signal_spectrum) derives I_cm via the Hockanson 1996 model:

I_cm(f) = (2 * L_gnd / (L_cable_per_m * cable_length)) * I_signal(f)

The omega in V_gb cancels the omega in cable Z_cm so the ratio is frequency-independent in the band of interest. Explicit cm_current_a still wins if set.

2. Montrose envelope spectrum

The existing harmonic_magnitude lands on the exact sinc curve, which has deep nulls at integer f*tau. Real chambers don't see those nulls -- EMI receivers integrate over their IF bandwidth and real signals have edge jitter. Added envelope_magnitude and envelope_sweep that follow the Bode-style two-corner envelope:

|I_env(f)| = 2 * I_peak * d * min(1, 1/(pi*f*tau)) * min(1, 1/(pi*f*t_r))

analyze_board now uses envelope_sweep. This is the standard pre-compliance form (Montrose Ch 5).

3. Cables in AnalysisConfig + CLI

AnalysisConfig::cables is a new vector. analyze_board power-sums loop + cable contributions per frequency, returns per-cable diagnostic in AnalysisResult::cables. CLI gains --cable-length-cm, --ground-inductance-nh, --cable-cm-uh-per-m, --cable-cm-ua.

TI re-validation -- new numbers

Same fixture (30 mm SCLK over 0.2 mm GND, 6 mA / 2 ns CMOS) plus:

• Cable: 30 cm USB, 1.0 uH/m, L_gnd = 15 nH (mid-range "real digital board", not per-test fitted)

Test	Measured	emikit	Gap
1 (SCLK=10 MHz, 9.7 kSPS)	34.67	31.64	-3.0 dB
2 (SCLK=50 MHz, 769 kSPS)	54.73	47.61	-7.1 dB
3 (SCLK=10 MHz, 232 kSPS)	51.06	33.63	-17.4 dB

Tests 1 and 2 within pre-compliance tolerance. Test 3 is the honest residual -- same SCLK as Test 1 but 17 dB more chamber emission because of higher sample-rate activity. The single-trace continuous-clock model doesn't see that. Documented as a known limitation in VALIDATION.md.

tests:
114 assertions / 44 cases. 20 new across cable_cm_test.cpp (estimator math + TI working point) and analysis_test.cpp (cable wired into BoardAnalysis, mixed loop+cable verdict, cable-only NoData semantics).

Known follow-ups

• Activity-factor / multi-net summing is the obvious next step to close the Test 3 gap. emikit currently models one trace at one clock.
• The 15 nH L_gnd is engineering judgement, not a measurement. Could be made empirical by feeding back from pdnkit's plane impedance once that ships.
• More reference pairs. Hockanson's 1996 paper has its own test boards with cleaner geometry.