[rfc] Add builtin_type_<floatformat>

[Andrew Cagney]

Hello,

(Almost there)

The attached patch adds a builtin_type for each of the machine dependant 
float formats.

The intent is that, like the builtin_type_intNN already used by MIPS, 
these make it possible for a target to exactly describe their 
floating-point registers as they appear in a buffer.

At present several things happen:

o 
builtin_type_long_double gets used and
	the target programmer hopes that this
	matches the ABI (typically it does but
	sometimes it doesn't!)

	Instead something like builtin_type_i387_ext
	can be specified which exactly describes
	the register buffer.

o 
GDB does lossy target-fp -> DOUBLEST ->
	??? conversions when displaying FP registers
	(CONVERT_TO_VIRTUAL) and this is no longer
	required.  REGISTER_VIRTUAL_TYPE can
	byte-for-byte, describe the
	buffer being returned.

	Just need someone to fix the code that
	prints such types :-)

Thoughts?

The first version of this function and 
builtin_type_floatformat_<floatformat> but I dropped that (for no real 
reason other than it was shorter).

	Andrew