[P3] Support flit for NoC pkg

[tancheng]

There are two NoCs now, one for inter-CGRAs, and one for intra-CGRA (i.e., inter-tiles for delivering ctrl/const).

Instead of extending the pkg format for any novel signal, we can make one pkg be multi-flit, and fix the bandwidth. This would increase the complexity of the controller (requiring additional serializer/deserializer or assembler/disassembler) but very necessary.