Tidy up ropp_fm_abel

[Ian Culverwell]

The ropp_fm_abel*.f90 routines could do with a tidy:

• Define the various constants (1.0, 1.0e-6, 10.0, 0.157 etc) as real parameters;

• Replace the 0.5-SIGN(0.5, x) cleverness with simple IF statements;

• Maybe calculate the Abel integrand in one go, using f90 array manipulations, rather than nested loops - this might also allow the erf calculations to be hived off to an internal/external subroutine without clobbering the performance;

• Examine safer ways of estimating int_up - int_low when they're very close (as they sometimes are).

And anything else that comes to mind.

Similar remarks apply to ropp_pp_{abel/invert}_{lin/exp}.f90.