patch for insn size > word size

[Doug Evans]

Stan Cox writes:
 > This proposed patch adds support for architectures where the insn size
 > is larger than the word size.  The cgen changes adds
 > CGEN_CPU_WORD_SIZE so it is easy to get this value from C code and
 > makes insn_value an array.  The opcode changes read, print, and insert
 > each insn word individually.  The code assumes if an insn is larger
 > than the word size, then it is twice as large.  Other possibilities are:
 > 1. check that insn size is twice word size
 > 2. generalize for an insn that is any multiple of word size
 >    (seems like it may not be worth the effort)
 > Also tested on an existing port where CGEN_INT_INSN_P is 0

I'd like to think about this one for a day or two.