patch: fix performance problems with repated decomprimation of varlena values in plpgsql

Hello

this patch remove a multiple detoasting of varlena values in plpgsql.

It is usable mainly for iteration over longer array directly loaded
from relation.

It's doesn't have a impact on semantic or behave - it's just eliminate
some performance trap.

sample: table 10000 rows one column with array with 1000 string fields:

patched pl time: 6 sec
unpatched pl time: 170 sec

This doesn't change my opinion on FOR-IN-ARRAY cycle (is still
important for readability) - just remove one critical performance
issue

Regards

Pavel Stehule