Neko milliseconds resolution

I'm sure I knew this at some point, but neko's milliseconds only has seconds resolution:

```haxe
import haxe.Timer;

function main() {
var t = Timer.milliseconds();
final tEnd = t + 5000;
while (true) {
final t2 = Timer.milliseconds();
if (t2 >= tEnd) {
break;
}
if (t != t2) {
trace('$t -> $t2');
t = t2;
}

}
trace(t, tEnd);
}
```

```
source/Main.hx:12: 1769515195000 -> 1769515196000
source/Main.hx:12: 1769515196000 -> 1769515197000
source/Main.hx:12: 1769515197000 -> 1769515198000
source/Main.hx:12: 1769515198000 -> 1769515199000
source/Main.hx:17: 1769515199000, 1769515200000
```

The main issue with this is that it can turn a `delay(1)` into a `delay(1000)` with some bad luck, which is what I assume happens on the Issue37 test.

Start by running the Haxe program in the issue and inspect Neko's implementation of haxe.Timer.milliseconds. Verify the timing behavior against the Issue37 test; done means millisecond readings have sub-second resolution so a delay(1) does not become delay(1000).

Written by the indexing model from the issue text.