Allow user defined pipelines of fixed latency

[JulianKemmerer]

Consider this pipeline...

void frame_buffer_display()
{
  vga_signals_t vga_signals = vga_timing();

  pixel_t vga_pixel = user_func(vga_signals);
  
  pmod_register_outputs(vga_signals, color);
}

Right now if the user's user_func is autopipelined then functionality is maintained.

But if the user has hand coded user_func with some other fixed delay (ex. maybe 1 or 2 cycle BRAM latency...) then auto pipelining does not work. The function is seen as combinatorial logic, zero latency, like a regular function using static local state vars etc...

Instead allow users to pragma specify the fixed latency they have coded. Ex. maybe like...

#pragma FUNC_LATENCY user_func 2
pixel_t user_func(vga_signals_t vga)
...

Which would do the proper 2 cycle delaying of signals in the places where user_func is used...

Makes the --comb flag for 'everything comb logic' mean something different...