Changing how Lambdas are normalized if Simplex=True

[facero]

When testing the IAE and the BSP on X-ray spectra, the BSP was giving strange spectra in ~5-10% of the cases.
Investigation revealed that this was due to the simplex constraint not being followed (Sum(Lambdas)>1) hence producing strange spectra as an output.

This PR modifies two things:

• Lambdas need to be divided by sum(L) not sum(abs(L))
• new Lambdas were not updated in params["Lambda"]

When testing Sum(L) is now 1 for the cases I tested and BSP and FastInterp now provide consistent results.

[facero]

Note that in the interpolator the absolute value is not present which would explain why FI was working and not BSP

IAE/codes/IAE_JAX_v2_devl.py

Line 534 in 2873e3e

Lambda = Lambda / (np.sum(Lambda, axis=1)[:, np.newaxis] + 1e-3) # not really a projection on the simplex

However a similar issue could exist here in the sparse_code interpolator where the absolute value is present. Maybe its worth fixing it as well.

IAE/codes/IAE_JAX_v2_devl.py

Line 529 in 2873e3e

Lambda = Lambda / (np.sum(np.abs(Lambda), axis=1)[:, np.newaxis] + 1e-3) # not really a projection on the simplex