Convert input tensor shape from `[ne03, ne02, n, k]` to `[ne03 * x, ne02 * y, n, k]` ~Add a [pad](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/MasterOpDef.html#pad) node or [concat](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/MasterOpDef.html#concat) node before the mulmat~ Found that its more similar to `gather` or `resize`: [Gather -QNN](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/MasterOpDef.html#gather), [Resize - QNN](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/MasterOpDef.html#resize), see also: [ggml-cpu.c#L7389](https://github.com/ggerganov/llama.cpp/blob/18429220bdb344da1bc7df9bc580c7b41b3cd57b/ggml/src/ggml-cpu/ggml-cpu.c#L7389)