Run 4 perf tests instead of 2

[j-f1]

It looks like the perf test script is running in ~2m before this change, which leaves plenty of time to gather more samples to hopefully smooth out the results so perf changes can be better caught.

I changed it to run the perf tests four times, with 2 in parallel at once. The constants can be easily changed if we find it’s still not stable enough.

[github-actions]

Time Change: -112.75ms (0%)

Total Time: 14,525.25ms

ℹ️ View Unchanged

Test name	Duration	Change
Serialization/Write JavaScript number directly	184ms	+1.5ms (0%)
Serialization/Write JavaScript string directly	184ms	0ms
Serialization/Swift Int to JavaScript	5,247.25ms	-157.25ms (2%)
Serialization/Swift String to JavaScript	5,377.75ms	-32.75ms (0%)
Object heap/Increment and decrement RC	3,532.25ms	+75.75ms (2%)

_{performance-action}

[MaxDesiatov]

0% here is good. I guess we'll need to see it running on more PRs to understand if the variance is really lower with 4 runs instead of 2.