3.x: fix switchMaps inconsistency swallowing errors when cancelled

[akarnokd]

The various switchMap operators did not report any accumulated exceptions to the global error handler when the sequence was cancelled (and thus the errors would never be delivered through the regular channels).

[codecov]

Codecov Report

Merging #6572 into 3.x will increase coverage by <.01%.
The diff coverage is 100%.

@@             Coverage Diff              @@
##                3.x    #6572      +/-   ##
============================================
+ Coverage     98.17%   98.18%   +<.01%     
- Complexity     6178     6182       +4     
============================================
  Files           678      678              
  Lines         44886    44896      +10     
  Branches       6212     6211       -1     
============================================
+ Hits          44069    44083      +14     
+ Misses          273      272       -1     
+ Partials        544      541       -3

Impacted Files	Coverage Δ	Complexity Δ
...internal/operators/flowable/FlowableSwitchMap.java	97.65% <100%> (+0.95%)	3 <0> (ø)	⬇️
...rnal/operators/observable/ObservableSwitchMap.java	94.7% <100%> (+0.02%)	3 <0> (ø)	⬇️
...ternal/operators/observable/ObservableFlatMap.java	89.25% <100%> (-1.69%)	3 <0> (ø)
...ternal/operators/mixed/FlowableSwitchMapMaybe.java	100% <100%> (ø)	2 <0> (ø)	⬇️
.../operators/mixed/FlowableSwitchMapCompletable.java	100% <100%> (ø)	2 <0> (ø)	⬇️
...x/internal/operators/flowable/FlowableFlatMap.java	89.94% <100%> (+0.47%)	4 <0> (ø)	⬇️
...rnal/operators/mixed/ObservableSwitchMapMaybe.java	100% <100%> (ø)	3 <0> (ø)	⬇️
...ernal/operators/mixed/FlowableSwitchMapSingle.java	100% <100%> (ø)	2 <0> (ø)	⬇️
...perators/mixed/ObservableSwitchMapCompletable.java	100% <100%> (+1.05%)	3 <0> (ø)	⬇️
...va/io/reactivex/internal/util/AtomicThrowable.java	100% <100%> (ø)	8 <3> (+3)	⬆️
... and 36 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 aa8e03f...61a5acd. Read the comment docs.

[vanniktech]

Does 2.x also need a fix?

[akarnokd]

It is certainly an inconsistency and goes against the no-error lost principle introduced via 2.x. It was reported (#6571) in conjunction with 2.x but the changes introduce more places of undeliverable exceptions. I don't know what such change could mean for users of the 2.2.x line.