This makes the IR consistent with x86 which already uses the per-element
transfer size for the scatter/gather memory operand size.

Issues: #5365, #5036, #6561

Change the scalar+vector scatter/gather instructions to use a
per-element size for their memory operand and update the codec tests
accordingly.

Change the vector+immediate scatter/gather instructions to use a
per-element size for their memory operand and update the codec tests
accordingly.

Change the scalar+scalar predicated contiguous load/store instructions
to use a per-element size for their memory operand and update the codec
tests accordingly.

The codec changes made in the previous patch mean we can combine some
of the scalar+scalar encode/decode functions and simplify the codec.

Change the scalar+scalar replicating predicated contiguous load/store
instructions to use a per-element size for their memory operand and
update the codec tests accordingly.

Change the scalar+immediate predicated contiguous load/store instructions
to use a per-element size for their memory operand and update the codec
tests accordingly.

Update the documentation comments to reflect the IR changes.

Now that the IR uses the per-element size drx_expand_scatter_gather()
will need to calculate the full memory transfer size.

Now that the SVE scatter/gather and predicated contiguous load/store
instructions use per-element sizes we don't need this workaround
any more.