test: improve assertions in pummel/test-timers

[Trott]

• Timers should not fire early. Check for that.
• Allow the wiggle-room to increase on subsequent iterations of
  intervals.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• commit message follows commit guidelines

[Trott]

Pummel CI: https://ci.nodejs.org/job/node-test-commit-custom-suites-freestyle/16430/

[Trott]

This fixes the flakiness of the test on macOS in CI.

master branch, 71 failures in 200 runs: https://ci.nodejs.org/job/node-stress-single-test/175/

this PR, 0 failures in 200 runs: https://ci.nodejs.org/job/node-stress-single-test/176/

@nodejs/testing

[Trott]

@nodejs/timers

[Trott]

@nodejs/collaborators This could use reviews.

[Flarna]

As far as I know windows timers sometimes fire slightly earlier. At least I faced issues with flaky timer tests in an earlier project because of this.
It may depend on the timeout, the windows version and the load on the machine.

I did a fast test and at least for a 16 and a 30ms timeout I saw that callback was called 0.x ms earlier then expected.
I measured time using process.hrtime() not sure if this makes a difference.

[Trott]

As far as I know windows timers sometimes fire slightly earlier. At least I faced issues with flaky timer tests in an earlier project because of this.
It may depend on the timeout, the windows version and the load on the machine.

I did a fast test and at least for a 16 and a 30ms timeout I saw that callback was called 0.x ms earlier then expected.
I measured time using process.hrtime() not sure if this makes a difference.

Firing less than 1ms early based on process.hrtime() seems plausible to me due to rounding errors. I'm inclined to leave this as is and then allow for 1 millisecond early if it proves to be flaky.

[Trott]

Landed in 87c433e