clarify that time_ns is monotonic

[stevengj]

Since time_ns() is based on uv_hrtime, the docs should note that it is guaranteed to be monotonic (see also #2464 and this discourse thread). This PR also notes that you can't compare times across machines or reboots, and that the timing resolution is system-dependent.

On Unix systems, it calls clock_gettime(CLOCK_MONOTONIC, &t) and the GNU libc manual notes that the reference time "may change if the system is rebooted or suspended." On Windows, it calls QueryPerformanceCounter.

(This docstring was last discussed in #54696.)

Also, I changed

The primary use is for measuring the elapsed time between two moments in time.

To "The primary use is for measuring elapsed times during program execution" (emphasis added), since you don't want to use this to compare arbitrary "moments", e.g. from different runs of a program (that may span a system reboot).

[stevengj]

For reference, I also want link the discussion at w3c/hr-time#115 — it turns out that whether CLOCK_MONOTONIC ticks during sleep is system-dependent.

[PallHaraldsson]

This should be backported to older versions, at least latest 1.10.x, not just its next minor version?! Specifically, and also pointing out, and asking about general doc policy, when it's not a new feature, but applied to the existing one, like here.

[KristofferC]

Why is this in its own paragraph with parenthesis? Seems like it is just as relevant as the other text.

[stevengj]

Because I thought it was kind of obvious that units ≠ resolution (i.e. precision ≠ accuracy), so this paragraph is just a gentle reminder / clarification (since the preceding paragraph already says that the units are ns)? But if you want to remove the parens that's fine with me.