[Bug-openmcl] decode-float exception

[bryan o'connor]

the number comparison tests are causing an unhandled exception
in decode-float.  i think i've simplified it down to this form:

   (decode-float least-positive-double-float)

decode-float is getting the correct (or perhaps just consistent)
answer, but it seems to be corrupting something along the way.

? (trace decode-float)
NIL
? (decode-float least-positive-double-float)
  Calling (DECODE-FLOAT 5.0D-324)
  DECODE-FLOAT returned 3 values :
       0.5D0
       -1073
       1.0D0
Unhandled exception 11 at 0x00006b34, context->regs at #xf0135758
Write operation to unmapped address 0x432fffe0
? for help
[15529] OpenMCL kernel debugger:

% sbcl
[...]
* (decode-float 5.0D-324)
0.5d0
-1073
1.0d0