base64 decoding must be done per-chunk

[nburlett]

Individually-encoded base64 chunks must be decoded separately
due to padding at the end of base64-encoded chunks. Concatenating
base64 chunks and then decoding them may result in a different string
than decoding and then concatenating.

Lamson currently concatenates the base64 encoded chunks and then decodes them, potentially resulting in a different message than originally sent.

However, the characterset decoding into unicode must be done jointly,
since some mail clients will split the base64 encoding in the middle
of Unicode characters