Thunderbird mbox parsing: CRLF headers cause silent empty scans

When scanning a Thunderbird IMAP mbox cache in Python with a hand-rolled splitter, messages may use CRLF line endings, so finding the header/body boundary with b'\n\n' fails and date filters silently drop every message, returning no output and no error. Split on lines starting with 'From - ' and regex the Date header from the first few KB of raw bytes instead. Treat an empty result as a possible parser bug, and confirm with grep -a on Date lines before concluding the mail is absent. Also decode MIME parts before keyword matching, since many bodies are base64 and raw byte searches miss them.