Fix for converting notebooks that contain unicode characters.

[astrofrog]

I tried converting a notebook that had a ü in it using nbconvert, and get:

$ ipython3 nbconvert *.ipynb --to latex --post PDF
[NbConvertApp] Using existing profile dir: '/Users/tom/.ipython/profile_default'
[NbConvertApp] Converting notebook 00. About the course.ipynb to latex
[NbConvertApp] Support files will be in 00. About the course_files/
[NbConvertApp] Loaded template latex_article.tplx
[NbConvertApp] Writing 17249 bytes to ./00. About the course.tex
b'stitut f\xc3\xbcr\nAstronomi'
Traceback (most recent call last):
  File "/Users/tom/Library/Python/3.3/bin/ipython3", line 9, in <module>
    load_entry_point('ipython==1.0.0-dev', 'console_scripts', 'ipython3')()
  File "/Users/tom/Library/Python/3.3/lib/python/site-packages/ipython-1.0.0_dev-py3.3.egg/IPython/__init__.py", line 118, in start_ipython
    return launch_new_instance(argv=argv, **kwargs)
  File "/Users/tom/Library/Python/3.3/lib/python/site-packages/ipython-1.0.0_dev-py3.3.egg/IPython/config/application.py", line 539, in launch_instance
    app.start()
  File "/Users/tom/Library/Python/3.3/lib/python/site-packages/ipython-1.0.0_dev-py3.3.egg/IPython/terminal/ipapp.py", line 355, in start
    return self.subapp.start()
  File "/Users/tom/Library/Python/3.3/lib/python/site-packages/ipython-1.0.0_dev-py3.3.egg/IPython/nbconvert/nbconvertapp.py", line 266, in start
    self.convert_notebooks()
  File "/Users/tom/Library/Python/3.3/lib/python/site-packages/ipython-1.0.0_dev-py3.3.egg/IPython/nbconvert/nbconvertapp.py", line 305, in convert_notebooks
    write_resultes = self.writer.write(output, resources, notebook_name=notebook_name)
  File "/Users/tom/Library/Python/3.3/lib/python/site-packages/ipython-1.0.0_dev-py3.3.egg/IPython/nbconvert/writers/files.py", line 109, in write
    f.write(output)
UnicodeEncodeError: 'ascii' codec can't encode character '\xfc' in position 11338: ordinal not in range(128)

The attached code seems to fix this, but understanding the intricacies of unicode isn't my strongest point, so let me know if it's the completely wrong approach :)

[takluyver]

We'll need to check whether output is always unicode, or if there's some way that it can be a byte string.

Also, it would be slightly neater to use with io.open(... encoding='utf-8') instead of the explicit .encode() call.

[astrofrog]

@takluyver - thanks, I'll give those suggestions a try

[minrk]

output should always be unicode

[minrk]

Go ahead make make the io.open(...encoding='utf-8') change, and this should be good to go.

[minrk]

@astrofrog want to make that change?

[minrk]

Merged here and applied finishing touches in #3879.

[astrofrog]

@minrk - thanks for finishing this up, I was traveling today and didn't get a chance to do this sooner.

[minrk]

No problem - sorry for my impatience, just getting ready to cut the release candidate.