SAT-2978 - SDC Series binops corrupt operands if fill_value is used

[kozlov-alexey]

This PR solves multiple issues in implementation of Series arithmetic and comparison methods:

• fixes wrong filling behavior (e.g. np.nan + np.nan was 2 * fill_value)
• removes inplace=True operations that corrupt operands
• fixes broken scalar + series use cases
• fixes a sub case of common case when series indexes are equal, and hence no alignment should happen
• adds performance tests for all different implementations
• moves operators and binop methods unit tests to a separate test suite
• unskips many tests incorrectly decorated with skip_parallel

[pep8speaks]

Hello @kozlov-alexey! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! :beers:

Comment last updated at 2020-07-27 14:09:22 UTC

[kozlov-alexey]

Performance results show that new impl behaves similar to the old one, e.g. for a case of adding two equal size series with none indexes: on master (df4d8d3459be7dff3) name	nthreads	type	size	median	min	max	compile	boxing
Series.add_series_same_size	1	Python	60000000	0.88369	0.882879	0.891595
	1	SDC	60000000	0.368971	0.368446	0.36941	0.439162	0.000547
	2	SDC	60000000	0.273992	0.273375	0.299651	0.4844	0.000718
	4	SDC	60000000	0.141599	0.141549	0.142013	0.471689	0.000618
	8	SDC	60000000	0.073205	0.073156	0.073496	0.471056	0.000618
	16	SDC	60000000	0.040385	0.040329	0.044085	0.473571	0.000611

with #891 name	nthreads	type	size	median	min	max	compile	boxing
Series.add_series_same_size	1	Python	60000000	1.170385	1.167961	1.171951
	1	SDC	60000000	0.349359	0.349062	0.350519	0.162805	0.000558
	2	SDC	60000000	0.233585	0.23304	0.23491	0.232838	0.000659
	4	SDC	60000000	0.122814	0.122562	0.122925	0.227218	0.000718
	8	SDC	60000000	0.061027	0.060741	0.061105	0.207034	0.000677
	16	SDC	60000000	0.032536	0.032402	0.033507	0.206282	0.000671

There's a small degradation in absolute values for series + scalar case, but scalability is not affected. For common case and it's equal indexes sub-case, exec times were improved (previously these test just hang). See https://jira.devtools.intel.com/browse/SAT-2978 for details.