In the documentation for xml.sax.xmlreader.InputSource objects
(section 8.12.4 of the Library Reference) we find that
users of InputSource objects should use the following
sequence to get their input data:

1. If the InputSource has a character stream, use that.
2. Otherwise, if the InputSource has a byte stream, use that.
3. Otherwise, open a URI connection to the system ID.

The parse() method of IncrementalParser skips step 1.

In addition, we need to add a method
getSourceEncoding() to the XMLReader interface;
if non-null, it will indicate to the parser that
the input is a byte stream in the given encoding.

The documentation should indicate what the parser
should do if the XML itself announces that its
encoding is something else. I propose that the parser should
be required to raise an error in that case.

See also #1483.