[musl] accuracy

[Paul Zimmermann <Paul.Zimmermann@inria.fr>]

       Hi,

with the two patches from Szabolcs on top of musl-1.2.2, I confirm the results
for acosh and sinh are back to normal for the whole binary32 range:

Checking function acosh with MPFR_RNDN
...
libm wrong by up to 2.01e+00 ulp(s) [2] for x=0x1.01cb98p+0
acoshf     gives 0x1.e4ce84p-4
mpfr_acosh gives 0x1.e4ce8p-4
Total: errors=295917691 (6.92%) errors2=2708 maxerr=2.01e+00 ulp(s)

Checking function sinh with MPFR_RNDN
...
libm wrong by up to 1.83e+00 ulp(s) [2] for x=0x1.783ab6p-1
sinhf     gives 0x1.9b038p-1
mpfr_sinh gives 0x1.9b0384p-1
Total: errors=72811936 (1.70%) errors2=31216 maxerr=1.83e+00 ulp(s)

Paul