nan_to_num in process.regression leaves inf as 1.8e308 #122

ridge_regression_fast calls np.nan_to_num(deltaf_series, nan=0.0) (regression.py:75), which maps NaN to 0
but leaves the default posinf/neginf handling in place, so ±inf becomes ±1.8e308 rather than 0. That value
overflows in the matmuls below it, giving inf coefficients.

∆F/F produces inf from the same zero-baseline division that produces its NaNs: (F - F0) / F0 with F0 == 0
and F != 0.

_pixel_ridge_regression (regression.py:221) has the same call.

Suggested fix

Pass posinf=0.0, neginf=0.0 at both sites.