Feature/depthwise convolution

also set pe/simd for non-DW for faster results

Cycle estimation annotation transform

Feature/expected cycles per layer

Merge branch 'feature/expected_cycles_per_layer' of https://github.com/Xilinx/finn into feature/expected_cycles_per_layer

Added transform for building with custom (non-Pynq) Zynq shell

Fixes Xilinx/finn#205

among others this will give correct TLastMarker behavior