[Feature] CDF-based dataset handling

[melekderman]

Is your feature request related to a problem? Please describe.

DistributionMultiTable could only be constructed from a tabulated PDF; the CDF was always derived internally via multi_cdf_from_pdf. Electron reaction data files (ionization secondary-electron energy distributions and elastic-scattering cosine) are provided as a tabulated CDF. There was no way to feed this data into the multi-table sampler, and degenerate single-point tables / zero-denominator cases were not handled, which could cause divide-by-zero or NaN samples.

Describe the feature you'd like.

Add a "direct-CDF" mode to DistributionMultiTable and the multi-table
sampler:

• DistributionMultiTable accepts an optional cdf= argument; when given,
  the CDF is used directly and PDF derivation is skipped.
• ElectronReactionIonization / ElectronReactionElasticScattering detect a
  CDF dataset in the HDF5 group and use it, falling back to PDF otherwise.
• _sample_multi_table handles three cases:
  • degenerate table (size == 1) → return the single value,
  • direct-CDF mode (pdf_length == 0) → linear interpolation on the CDF,
  • PDF-based mode (neutron data) → existing sampling, guarded against the p0 == p1 and c0 == c1 zero-denominator cases.

This keeps full backward compatibility for neutrons while enabling CDF sampling from electron data.

How can this issue be closed?

• DistributionMultiTable supports direct CDF input
• Electron ionization & elastic-scattering reactions read CDF data
• _sample_multi_table samples direct-CDF and PDF-based tables without divide-by-zero / NaN
• Existing neutron sampling results unchanged (regression check)