Changes from 1.4:

- API CHANGE: On Windows, AIR_EXISTS expands to airExist_d(), which
may be slow, because it uses 64-bit integer bit-masking operations,
but we'll have to find out.  Non-Windows: use !((x)-(x)).
- API CHANGE: airSanity() makes sure that sizeof(float)==sizeof(int)==4
and sizeof(double)==sizeof(long long)==8, in order to justify the
bitmasking behind AIR_EXISTS_F() and AIR_EXISTS_D().
- API NEW: added AIR_EXISTS_F(), a macro for testing non-specialness which
uses bit masking instead of the !((x)-(x)) trick.  Is therefore robust 
against all stripes of optimizing compilers, but unfortunately can ONLY
take a "float" as argument.  Basis of airExists_f().
- API NEW: added AIR_EXISTS_D(), like AIR_EXISTS_F(), but for doubles.
Basis of airExists_d().
- API NEW: airFPGen_d(), airFPClass_d(), airFPFprintf_d()
- API NEW: airFPPartsToVal_d(), airFPValToParts_d().
- API NEW: added airMy32Bit and AIR_32BIT
- API NEW: added AIR_SIZE_T_FMT which is the format string you need to
print something of type size_t.  But you need to compile with 
TEEM_32BIT defined to 0 or 1

754.c:
- NEW: _airDouble allows accessing a double as a double, as one
airULLong, as two unsigned ints, or as the three fields (sign,
exp, frac).  These are used in airFPFprintf_d().