[Math] Validate Diode-Fit accuracy and optimizer convergence

[MasterUser43]

Problem

Initial testing of the single-diode model fitting (diode_fit_resistances) in PR #10 showed significant numerical discrepancies:

• Synthetic Data: The optimizer landed exactly on the initial guess (Rs=20, Rsh=1e5), failing to converge to the known ground truth.
• Dark Scan Data: Disagreed with the derivative_resistances method by 18x to 60x on the same curve.

Investigation Steps

1. Residual Scaling: Re-evaluate the tanh soft-clipping and f_scale=1e-6 parameters in the least-squares call. They may be flattening the gradient too aggressively for low-signal curves.
2. Signal-to-Noise: Characterize the math against an Illuminated Sweep (Hal-320 solar sim). The current dark-scan testing ($J_{sc} \approx 0.056 , \text{mA/cm}^2$) may simply lack the signal strength required for a 5-parameter fit.
3. Provisional UI Flag: Until validated, the GUI should likely display these values with a "Provisional" or "Beta" label to prevent reliance on unverified data.

[MasterUser43]

f_scale = max(0.01 * np.max(np.abs(I)), 1e-12)

Might be too sensitive to noise. Although math improvements were observed, it seems like further work needs to be done.

Rsh for diode seems to be accurate and Rs for derivative also seems to be accurate.

Their respective pairs seem to diverge too heavy