Improve examples in docstrings

Pull request #433 brings to mind some things that could be improved in docstrings more generally. Cleaning these up might be a good starter issue.

1. Obviously, code examples should be correct and actually work. :)

2. Code examples should also be self-contained if possible, and avoid referring to files in sample-data (most people reading the documentation are probably not running the code directly from the git repository.) In most cases, it's better to use examples from published databases on PhysioNet.

3. Code examples should be written in a consistent style. We generally use numpy style (https://numpydoc.readthedocs.io/en/latest/format.html) for docstrings, but we haven't been entirely consistent. In particular, numpy style requires examples to be written in "doctest" style, like this:

    """
    Examples
    --------
    >>> np.add([[1, 2], [3, 4]],
    ...        [[5, 6], [7, 8]])
    array([[ 6,  8],
           [10, 12]])
    """

This format has some nice properties (it can be tested automatically by doctest, and formatted nicely by sphinx.) On the other hand, the "..." is a little annoying when you want to copy and paste an example.

Start by reviewing pull request #433 and the existing docstrings it discusses. Identify examples that use sample-data or fail to follow the stated NumPy-style doctest format; done means the selected examples work, are self-contained where possible, use published PhysioNet data when appropriate, and follow a consistent style.