Add integration and regression tests

[menshikh-iv]

Right now, for testing smart_open (local file part) - it is not enough to run only tests from smart_open, needs to run gensim tests with smart_open from master (so I caught a bug today - hotfix #150).

Need to look at gensim usage of smart_open and add tests according to this behavior (better - something more "general" for all readers/writers, because the tests are rather monotonous).

CC: @mpenkov

[menshikh-iv]

What's need to implement

General

• S3

  • Run all tests with real amazon S3
  • Reconfigure Travis, add job for amazon (think about distinct account, encrypt API keys, etc)
  • More tests for Amazon

• HDFS

  • Run all tests with hdfs (http + naitive)
  • Reconfigure Travis, add job (probably with hdfs docker)

• HTTP

  • Run http tests with real server
  • Re-configure Travis

Regression

• Write regression tests (for all stuff)
  • HDFS
  • S3
  • HTTP
• Store result always (need for investigation), probably on amazon / any cloud storage
• Auxiliary stuff (plot graphic, etc)

Gensim tests

• Go through gensim tests (probably corpora), look to use-cases, add more tests for smart_open

[menshikh-iv]

UPD

• Regression & integration for S3 done in [WIP] Add integration testing with real S3. Partial fix #151, #156 #158
• Integration for HTTP done in Refactoring smart_open to share compression and encoding functionality #185