Properly account for padding in PUTARG_SPLIT with GT_FIELD_LIST (#57279)

Properly account for padding in PUTARG_SPLIT with GT_FIELD_LIST (#57279)

We were writing the stack part to the outgoing stack area by a loop
incrementing the offset by the written size in each iteration. However,
it is possible due to padding and optimization that the written size is
smaller than the delta to the next field. Instead, use the offset of
each field minus the offset of the first stack field to find the offset
in the outgoing arg area.

Fix #57064