s390: Fix VFMA, VFMS, VFNMA, and VFNMS for 128-bit FP

s390: Fix VFMA, VFMS, VFNMA, and VFNMS for 128-bit FP

When trying to execute any of the vector float multiply-and-add family
instructions, Valgrind panics with the message "Iex_Qop with F128 data".

So far all F128 operations were implemented with instructions that operate
on FP register pairs.  However, such instructions don't exist for
Iop_MAddF128 and Iop_MSubF128, so they must be implemented with vector
instructions instead.  This was missed when emitting them in
"guest_s390_toIR.c".

Add the missing support.  This also involves adding a few new features to
host_s390_defs:

* a new vector operation S390_VEC_INIT_FROM_FPRS

* the capability to move the left half of a VR to an FPR

* S390_VEC_FLOAT_MADD/_MSUB for 128-bit sizes