[HandshakeToHW] Tracking issue

[mortbopet]

This is a tracking issue for the reimplementation of HandshakeToFIRRTL to a HW/ESI-based implementation:

• Operators
  • Mux
  • Conditional branch
  • Instance
  • Buffer
    • Seq
    • Fifo: A fifo is most easily implemented when some memory abstraction can be targeted (ready/write pointers, transparency, ...). We currently have seq.hlmem but it would be nice to also have a seq.fifo operation (which could lower to a seq.hlmem implementation).
  • Constant
  • Fork
  • Join
  • LazyFork
  • Load
  • Store
  • Memory
    • Ideally should lower to a seq.hlmem operation, see [Seq] seq.hlmem ODS and rationale #3973
  • External memory
  • Select
  • Sync
  • Pack
    • Partially done - lowerings exist but semi-blocked by [ExportVerilog] Struct assignment patterns not supported for all tools #4043
  • Unpack
• Support pruning i0 logic
  • HandshakeToHW attempts to to keep special-case logic for handling none (soon to be i0)-typed values. This implies that any handshake SSA value which is i0 still carries ready/valid/data semantics, however the data part of the bundle is of zero width. Given this style of lowering, we need a pass that is able to prune the i0 logic post-conversion (or keep it around until emission and let the emitter decide what to do with it).
• Figure out how to support memories
  • This will be ESI based, and should use an ESI service.
• Move integration tests from FIRRTL to HW implementation
• Remove HandshakeToFIRRTL