Re: Introducing --gc:arc

[cblake]

On Linux 5.4.2, `gcc 9.2.0 -march=native -O3 -fno-strict-aliasing` , head of 
devel (...ea39524bac14f6c89de5213509f4099) with a similarly tweaked run-5-times 
variant (using `epochTime` instead of `getTicks`), I see no discrepancy in 
elapsed time in `--gc:arc -d:release` and `--gc:arc -d:danger`. It does seem 
maybe more variable with only `-d:release`, but I have not studied it very 
systematically. I get numbers like 0.07 milliseconds per 1 loop of the 5 both 
ways..min == max, even. It does vary a lot from run-to-run of the program 
though, with both release & danger modes. I suspect we need a more reliable way 
to benchmark before being too worried.