x87 floating point immediates should be double precision

DR's decoder uses single-precision floats for the x87 immediates from instructions like `fldl2t`. However, these instructions push double extended-precision format values for the 80-bit x87. It seems best for DR to represent them as `double` rather than `float`. (I don't know that it's worth trying to use `long double` or something for a 10-byte format where supported.)

Xref #4408 where for AArch64 we are adding `double` to `opnd_t`.