lora conv1d replacement (#16643)
in LoRA code, it will use conv1d to do projection for qkv, while the
conv1d calculation is mathematically equivalent to matmul, and matmul is
much faster than conv1d.
The subsitution of the graph optimizer is: 1 conv1d >> 2 split + 1
squeeze + group_num matmul + 1 concat
with this optimizer, we see 10%+ in one 1P model
