2.x: Fix Observable.switchMap main onError not disposing the current inner source

[akarnokd]

The Observable.switchMap had bad logic in its main onError handler which didn't dispose the current inner Observable.

The Flowable version didn't have this logic error. Both variants received an unit test to verify the correct behavior.

Fixes #5832.

[codecov]

Codecov Report

Merging #5833 into 2.x will increase coverage by 0.01%.
The diff coverage is 100%.

@@             Coverage Diff              @@
##                2.x    #5833      +/-   ##
============================================
+ Coverage     96.42%   96.44%   +0.01%     
+ Complexity     5816     5815       -1     
============================================
  Files           634      634              
  Lines         41761    41760       -1     
  Branches       5796     5796              
============================================
+ Hits          40268    40275       +7     
+ Misses          578      573       -5     
+ Partials        915      912       -3

Impacted Files	Coverage Δ	Complexity Δ
...rnal/operators/observable/ObservableSwitchMap.java	90.62% <100%> (+1.8%)	3 <0> (ø)	⬇️
.../operators/completable/CompletableConcatArray.java	93.75% <0%> (-6.25%)	2% <0%> (ø)
.../operators/flowable/FlowableBlockingSubscribe.java	91.89% <0%> (-5.41%)	9% <0%> (-1%)
.../internal/operators/flowable/FlowableInterval.java	94.44% <0%> (-2.78%)	3% <0%> (ø)
...activex/internal/observers/QueueDrainObserver.java	61.53% <0%> (-2.57%)	12% <0%> (-1%)
...va/io/reactivex/internal/queue/SpscArrayQueue.java	97.61% <0%> (-2.39%)	22% <0%> (-1%)
...l/operators/observable/ObservableFlatMapMaybe.java	93.46% <0%> (-1.97%)	2% <0%> (ø)
...x/internal/operators/flowable/FlowablePublish.java	93.83% <0%> (-1.77%)	10% <0%> (ø)
...a/io/reactivex/internal/util/QueueDrainHelper.java	61.8% <0%> (-1.39%)	35% <0%> (-1%)
...ternal/operators/observable/ObservablePublish.java	95.61% <0%> (-0.88%)	10% <0%> (-1%)
... and 21 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 1fbc44f...cfa17ae. Read the comment docs.

[artem-zinnatullin]

Codecov indicates that some of possible variants are not tested, do you want to cover them in this PR?

[artem-zinnatullin]

It's not that I think code in the patch is broken, but that if we had test that was trying to go into onError when it's already done == true, we would have a hint that original code had some problems

[akarnokd]

That line is tested but parallelism on the CI is flaky and may not produce the necessary interleaving there.

The upstream may onError after a crash in the mapper thus we have to route that into the global error handler.

[artem-zinnatullin]

Can't we test it just by breaking RS protocol with main.onComplete(); main.onError()?

I might miss something, reviewing on mobile

[akarnokd]

This and this tests should cover that line. Apparently, the test did not involve ps2 so the inner source was not involved. Updated the unit tests accordingly.