1. in the language table 0x80 is not part of 8859-n and the range
0xa0-0xff is valid latin1.

2. don't call chartorune() first -- it is difficult to tell a bad rune from
ascii/latin1/binary.  deal with 4-byte utf-8.

3. added utf-(16|32)(be|le) detection (when byteorder marks are present).

4. support image bit-depths of 1, 2 and 4.

5. font file parsing now insists on newlines between entries and ignores
short lines.  subfonts that are split into two different images are also
properly located.

i wrote a seperate program chkfont that slowly checks most font parameters
and takes pains to complain in a meaningful way which is included.

6. iself's automatic cpu array was redeclared "static" so that non-explicitly
initialized locations would be 0.  added an array for elf type
(core file, relocatable, shared library, executable). 

less important changes
1. -h flag -- prints character histogram.  this is sometimes useful to
see why files lib /lib/glass are classified as binary.

2. added rustic support for detecting some dbcs character sets.
although the method is crude, we should not often be fooled on
real examples -- it's uncommon to have even-byte long runs of
latin1 characters.