charset.c.  The new DECODER framework is currently only used in
handler.c, and there in a horribly inefficient manner.  We should
use greater blocks of data, which would be much more efficient than
what we are currently doing.

Most of the other charset-related code still uses the old
mutt_display_char() &friends interface, which is actually ok as long
as you don't try to handle multibyte character sets.

The most notable change should be the one to mutt_get_translation():
It will delay the loading and parsing of character set information
files until it's really needed, catching a huge amount of standard
cases.  As a side effect, this will make "iso tagged as ascii"
"work" again, as long as both sides use the same iso character set.