Fix prescan bug: Second and invalid http-equiv

[openandclose]

#data
<meta http-equiv="refresh" http-equiv="Content-Type" content="text/html; charset=iso8859-2">
#encoding
None

The spec uses attribute list, to keep track of parsed attribute names,
discarding duplicates.

While html5lib._inputstream.EncodingParser returns: 'iso-8859-2'.

[openandclose]

This is the same bug as:
jsdom/html-encoding-sniffer#8 (fixed)