llama.cpp

fix(context): draft model fit vs load inconsistency

#25056

Merged