fix(VirtualTimeScheduler): rework flush so it won't lose actions

[baizulin]

Description:
Previously VirtualTimeScheduler.flush would lose the action that was on a verge of maxFrames limit which rendered testing of observables that are ticking indefinitely impossible. After the fix the user can set maxFrames, flush, make assertions and repeat the process as many times as needed.

Related issue:
None

[cartant]

Thanks for the PR. I will have a look at this tomorrow. (I also need to look at what's causing the dtslint run with TypeScript next to effect the errors in the Travis log - something that's unrelated to this PR.)

[baizulin]

@cartant Hi! Did you have a chance to take a look?

[cartant]

I looked at it briefly.

My concern is that maxFrames is no longer relevant given that its set to POSITIVE_INFINITY in TestScheduler#run. And that is the recommended mechanism for writing marble tests.

I'll wait to see what others have to say about it.

[benlesh]

Please review and make the suggested changes.

Thanks, @baizulin!

[benlesh]

@cartant ... VirtualTimeScheduler may be used on it's own, and this seems like a valid fix/bug to me.

[cartant]

LGTM

[baizulin]

@cartant @benlesh guys let's fix this bug already, it's been over a month

[benlesh]

@baizulin ... I appreciate that you put in the effort to submit this PR, I understand it's been a while and that can be frustrating. Please try to be respectful of our team members and our time, at this time all RxJS work is done by volunteers in our spare time.