This improves test coverage of the lexer. Before, a SpyReader's read function
would return pointers into a single string that contained the entire text. This
could have masked bugs where out-of-bounds characters were being read.
Now the chunks returned by the reader are copied into a separate buffer.