Batch the tokens of an expert in ggml-cpu's MoE path (patch 0022, +47 % prompt / +26 % generation for gpt-oss-20b on the Core Ultra module), switchable from Java

Batch the tokens of an expert in ggml-cpu's MoE path (patch 0022, +47 % prompt / +26 % generation for gpt-oss-20b on the Core Ultra module), switchable from Java #1284