[Feature] Separate hand claps from live recording

As we all know, when we record a live performance (in the audience) the applause always covers the music. Can AI be trained so that spleeter can separate an applause track? So the level of applause can be adjusted separately.

The issue names no files, tests, or entry points. Begin by defining the training data, model scope, and expected applause separation behavior; the work is complete only when applause can be separated and its level adjusted independently from the live recording.

Written by the indexing model from the issue text.