[Discussion] About the weight layout of Dense/BatchMatmul

Should we extend qnn.dense to qnn.matmul to handle the transpose in quantized models? We are facing similar issues in quantized bert models.

cc @comaniac @jcf94