Apply SF patch #1101726: Fix buffer overrun in tokenizer.c when a source file

with a PEP 263 encoding declaration results in long decoded line.