Handles the empty string encoding/decoding in Python 2.7

[mstfbl]

Additional fix to Issue #360
Fixes #366

Handles the case where, with Python 2.7, string s is valid with utf_8 encoding, but is the empty string. I discovered the need for this fix as 2 test cases that were skipped before PR #359 were failing even with PR #359, due to this case with the empty string on Python 2.7.

These tests were:

• TestNGramFeaturizer.test_ngramfeaturizer()
• TestLightGbmClassifier.test_lightgbmclassifier()

EDIT: My observation on the error provided in TestLightGbmClassifier.test_lightgbmclassifier() was incorrect. The error is only visible in TestNGramFeaturizer.test_ngramfeaturizer(). The exact failure is here and here, shown on the Mac 2.7 build.

[ganik]

    if (bp::extract<const char*>(str(s).encode("utf_8")).check())

What does this check() method do?

---

Refers to: src/NativeBridge/DataViewInterop.h:241 in 1f92b8a. [](commit_id = 1f92b8a, deletion_comment = False)

[mstfbl]

The check() method in this context is:

  template <class Ptr>
  inline bool extract_pointer<Ptr>::check() const
  {
      return m_source == Py_None || m_result != 0;
  }

According to the Boost docs here, I believe this functions checks whether the given object can be converted to a given C# type, which in this case would be <const char*>. This check is necessary to properly handle a given text from Python to C++, and later to C#.

[mstfbl]

As of commit d9b9544, the Mac Py2.7 build passes.

[mstfbl]

Closing this PR for now as we are temporarily disabling the corresponding tests for the Mac Python 2.7 builds with PR #391 .