Wire GPU NMF engine into CLI

[JerryIshihara]

Summary

This PR wires the existing standalone GPU NMF kernel into the cNMF CLI factorization path.

Changes

• Adds optional NMF engine CLI args for factorize
• Keeps the default path as CPU/sklearn behavior
• Enables GPU NMF only when --engine gpu is passed
• Passes GPU runtime options into the cNMF _nmf adapter
• Raises errors when GPU options are used with unsupported commands or without --engine gpu