Glb tiling for dwconv

[yuchen-mei]

1. update testbenches to support glb tiling, which runs the same kernel multiple times for conv with large number of channels
2. add end-to-end arg in garnet.py and write 0s to interconnect config regs in bitstream to avoid random flushes caused by previous kernels