Tap support in hwacc

[Kuree]

This PR adds tap stencil support in hwacc driver, tested with Halide-HLS codegen. Various improvements:

1. Add indexing in device tree overlay so that the driver will automatically arrange the channel orders to match with the pipeline_zynq call.
2. Use UBuffer to wrapper around tap values without resorting to cmabuffer. The user memory is set with UBuffer's id and stride (high and low), as these two values are meaningless in 1D tap values. If the height is set to 1, which is done automatically through f4graph+Halide, the driver will treat the UBuffer as tap values.
   • No extra changes needed in f4graph, which is an advantage of using tap values with Zynq CodeGen.
   • The only limitation is that only 1D taps are supported. However, if we have 2D taps, then we might as well just use 2D streaming interface (as an image).

Tested with seedark_hw in f4graph.

Need some sleep...