[LLHD] Flesh out zero-delay semantics

In various places it would be helpful if the llhd.drv operation (and potential future llhd.output) without any delay. This would allow implementing folds like:

%0 = llhd.prb %sigA
llhd.drv %sigB, %0
// or
%0 = llhd.prb %sigA
%sigB = llhd.output %0
// fold to
llhd.buffer %sigB, %sigA  // one-way connect from A to B

It's not quite clear what the simulation semantics of zero-delay should be. Simulation could just error out on zero-delays or replace them with a 1d delay. We might also choose to integrated the zero-delay signal into the event queue when updating the signal values in between time steps. Something similar is going to be needed for llhd.con anyway.

Start by examining the stated zero-delay cases for llhd.drv and potential llhd.output, then trace how llhd.buffer and llhd.con are expected to relate to them. Compare the proposed error, 1d replacement, and event-queue alternatives, and establish simulation behavior and folding criteria before implementation.