Allow code stubs to be language specific

[jhamrick]

Fixes #517

The idea here is that you can specify the code stubs and solution region delimeters for the ClearSolutions preprocessor as a dictionary keyed by language, where the language should be the same as the language of the kernelspec in the notebook metadata. This way, nbgrader can flexibly handle assignments with notebooks that have different kernels. For example, the config file might look like:

c = get_config()
c.ClearSolutions.code_stub = {
    "python": "# YOUR CODE HERE\nraise NotImplementedError()",
    "javascript": "// YOUR CODE HERE\nthrow new Error();"
}
c.ClearSolutions.begin_solution_delimeter = {
    "python": "### BEGIN SOLUTION",
    "javascript": "// BEGIN SOLUTION"
}
c.ClearSolutions.end_solution_delimeter = {
    "python": "### END SOLUTION",
    "javascript": "// END SOLUTION"
}

@randy3k does this seem like it would satisfy what you were looking for?