more work on test suite

Real bugs:
1. unicode_iconv needs to save the shift state when backing up
2. the init function for a converter has no idea if it is initializing
   for read or write.  should have two init functions

Treat malformed (and overlong, except maybe c0 80) UTF-8 characters
like 0xfffd.  Define for 0xfffd in converter header.

we need some more functions:
  a way to map character number to byte number
    (so we can continue to use strncmp)
  a strchr and strrchr using a unicode_char_t argument
  strcasecmp and strncasecmp

intermediate buffer:
  * how to size it?
  * should be circular for efficiency
  * should be bypassed when converting to/from UCS-4

more charsets
  Japanese charsets like JIS are very important
  These must be converted from Unicode tables using a script

name charsets appropriately
names of existing charsets should be changed.
eg ucs2 -> utf16
look at IANA registry

look at Solaris and glibc 2.1:
  make sure we use the same names for the same encodings
  see if we have to make it so our aliases apply even when
    using the system iconv (necessary when we want to use a fixed
    charset but existing systems don't agree on what it should be
    named)

if fromcode==tocode, then just copy through

merge with FriBidi

use gtk-doc

ucs readers which understand the feff convention

turn converters into shared objects of their own and dynamically load
them as requested

unicode_canonical_decomposition has a bad interface
Should let user give a buffer


================================================================

more properties:

* mirrored.  use 2-level table