fix(scheduler): getNow detection can randomly fail (fix #9632)

[felixbuenemann]

The previous detection code compared time stamps based on Date.now()
which are not monotonic, so the check could fail due to clock skew or
adjustments.

This fix changes the check to compare against performance.now() if it is
supported, because it is monotonic (strictly increasing).

What kind of change does this PR introduce? (check at least one)

• Bugfix
• Feature
• Code style update
• Refactor
• Build-related changes
• Other, please describe:

Does this PR introduce a breaking change? (check one)

• Yes
• No

If yes, please describe the impact and migration path for existing applications:

The PR fulfills these requirements:

• It's submitted to the dev branch for v2.x (or to a previous version branch), not the master branch
• When resolving a specific issue, it's referenced in the PR's title (e.g. fix #xxx[,#xxx], where "xxx" is the issue number)
• All tests are passing: https://github.com/vuejs/vue/blob/dev/.github/CONTRIBUTING.md#development-setup
• New/updated tests are included

If adding a new feature, the PR's description includes:

• A convincing reason for adding this feature (to avoid wasting your time, it's best to open a suggestion issue first and wait for approval before working on it)

Other information:

See the discussion in #9632 for additional info.

This needs to be backported to the 2.6 branch.

[yyx990803]

The check would lose the point if we compare the event timeStamp against a hi-res timeStamp. I think it should be good enough to only use performance.now when it's available since this only affects IE9.

[felixbuenemann]

The check would lose the point if we compare the event timeStamp against a hi-res timeStamp.

No, the point of the check is to see if the event timeStamp uses hi-res (like performance.now()) or lo-res (like Date.now()). So comparing if the event timeStamp is smaller or equal to performance.now() will work, because if it is lo-res it will be a lot bigger than performance.now().

Comparing to performance.now() fixes the problem, because it is not affected clock inaccuracies.

I think it should be good enough to only use performance.now when it's available since this only affects IE9.

No, it also affects IE 10/11 and any other browser that uses Unix timestamps (lo-res) for events.

[yyx990803]

You are right. Reviewing PRs with a jetlag is a bad idea.