Use clock_gettime for all performance timestamps

This MR adjusts the performance monitoring code to use `Process.clock_gettime` (thus `clock_gettime(3)`) instead of `Time.now`.

Using `Time.now` / `Time.new` adds more overhead than `Process.clock_gettime`, it also doesn't provide a way of getting timestamps in nanoseconds (which `Process.clock_gettime` does allow).

See merge request !4899
(cherry picked from commit 53ad9522)