[Seq] Unify `seq`-dialect memory representations

[mortbopet]

As noted in https://github.com/llvm/circt/pull/5009, https://github.com/llvm/circt/pull/5025, seq.firmem was introduced to initially help factoring out memory handling in the FIRRTL-to-HW pass. Once done, it is then intended that it and seq.hlmem can be unified into a single representation - seq.mem - seeing as they both essentially try to solve the same problem.

i.e.

• seq.firmem + seq.hlmem => seq.mem
• seq.firmem.read_port/write_port/read_write_port + seq.read/write/read_write => seq.read_port/write_port/read_write_port

[mortbopet]

CC @fabianschuiki

[fabianschuiki]

Thanks for opening a tracking issue!

[teqdruid]

@fabianschuiki How's this going?

[fabianschuiki]

Still letting the dust settle on seq.firmem internally. This has been enabled in the last release of firtool, and I'm collecting feedback and looking out for issues that come up with it in the next few release cycles -- just in case we need to quickly revert it. After that I'm all for merging these two representations, and homogenizing the latency annotations as needed.

[teqdruid]

@fabianschuiki certainly the dust has settled by now...

[fabianschuiki]

Yeah we should definitely unify these 😿