By importing unicode_literals from __future__ and casting to unicode at
two locations a simple case now works. However I have not audited the code
to find other locations where casts may be necessary. No new failures were
noted when running the testsuite.

New functions were introduced in openpy to deal with encoding in
python files. This commit removes redundant code from io and moves
source_to_unicode to openpy.

in namespace.py _inspect access was done on self instead of self.shell
fixing import of source_to_unicode in osm.py that became bad during rebase

fix to handle the case when len(lines)<2

when calling %run -d the filename available for pdb is <string> which is
not a proper filename. %run now adds a _exec_lines attribute containing the
correct filename.

Use read_py_file to get python source in unicode format.

Fix to interpret str docstrings with the encoding of the sourcefile.

Fix to interpret source of object with the encoding of the sourcefile.

Most files in linecache are stored with the filename as a bytestring
(py2.x). When calling getlines with a unicode filename there will be
implicit conversions using ascii that fails. This is avoided by an
explicit conversion using the filesystem encoding before calling
getlines.

For instance Windows IO error messages are localized and may contain
non-ascii chracters.