Subject: $+IO!<$+MO!~$+IO!<$+MO!~$+IO!<$+MO!~$+IO!<$+MO!~$+IO!<$+MO!~$+IO!<$+MO!~$+IO!<$+MO!~$+IO!<

Subject: =?iso-2022-cn-ext?b?GyQrSRtPITwbJCtNG08hfhskK0kbTyE8GyQrTRtPIX4bJCtJG08hPBskK00bTyF+GyQrSRtPITwbJCtNG08hfhskK0kbTyE8GyQrTRtPIX4bJCtJG08hPBskK00bTyF+GyQrSRtPITwbJCtNG08hfhskK0kbTyE8?=

No less than 191 characters!  Oh boy, this means that we'll have to
split the header for every 4 characters in some cases.

I've checked in a fix.  It

* Assumes a factor 8 worst case expansion for everything with B in
  `rfc2047-charset-encoding-alist' where I didn't know better.  It's
  true for iso-2022-cn-ext -- if someone knows better values for other
  charsets, please fill them in.

* Only fixes the 75 char limit on encoded-words.  It's still possible
  to exceed the 76 char line length limit, although less likely.