Add switch_map operator and equivalent starred and indexed

[giff-h]

This draws from the definition in rxjs and maintains parity with the
map operator and its variants

[coveralls]

Coverage increased (+0.09%) to 93.449% when pulling 7fbbd9d on hamstap85:feat/switch_map_operator into a227802 on ReactiveX:master.

[dbrattli]

Great work! The main building blocks of Rx is the merge functions (merge, switch_latest, concat) combined with the mapper forms the functions we know as flat_map (i.e map_merge), switch_map and concat_map. So we want to have switch_map as much as we want to have flat_map. We should also add concat_map.

[dbrattli]

Suggested change

	Project each element of an observable sequence into a new observable.
	Project each element of an observable sequence into a new observable. producing values only from the most
	recent observable sequence.

[dbrattli]

Suggested change

	by incorporating the element's index.
	by incorporating the element's index and producing values only from the most recent
	observable sequence.

[dbrattli]

Suggested change

	applied on unpacked elements as positional arguments.
	applied on unpacked elements as positional arguments, and only producing values only from the most recent
	observable sequence.

[dbrattli]

Same here: " ... producing values only from the most recent observable sequence.""" or else it looks like a flat_map.

[dbrattli]

@hamstap85 One thing that worries me about this PR is that the tests are derived from test_map and not e.g test_switchlatest. Thus the tests do not have the notion of time and behave more like a regular flat_map / concat_map in the sense that the switching property of switch_map is not tested. I.e the automatic disposal of the previous observable being subscribed. This is also reflected in the docstrings of the operators where the " producing values only from the most recent observable sequence" is missing. The current docstrings sounds more like a flat_map so I'm worried that your original impl also behaved more like a flat_map / concat_map. Sorry for being difficult, but I just need to be sure that you are implementing what you think you are implementing.

[giff-h]

@dbrattli No worries at all, thanks for the feedback.

What's the desired max line length in docstrings? Equal to python code max length enforced by black?

[dbrattli]

Docstrings should follow PEP-8, i.e 72 chars per line. PS: all the current tests for switch_map and switch_map_index pass if you replace switch_map with flat_map etc. The only difference is that switch_map uses of instead of identity as the default mapper function. But in my view they should either both use of or both use identity. Do you have a reference for another library using of for the default mapper? I'm not opposed for changing the default for flat_map but I cannot see how this is currently handled in e.g RxJS. FYI: https://www.baeldung.com/rxjava-flatmap-switchmap