* Remove assert(), since they can be done at cpp files if needed
* Calculate `(GTimer() >= disp_begin) && (GTimer() <= disp_end)` at the
  top level module only once, and wire such bool signal to where XSLog()
  is called. This can reduce the number of instances of counter created
  while GTimer() is callled.
* Remove xsLogLevel. It seems meaningless, since we either need all logs
  for debugging, or no logs for running tests only.
* With the above optimizion, running microbench with test input spends
  120s on 9900k with log completely disabled (comment out the log code),
  but only spends 147s on 9900k with log enabled.

ibuffer: fix io.in.ready to "!full"

ifu: fix pnpc vector in fetchPacket