optimize qd_compose_insert_string_iterator

Analysis of perf data for the receive path of a inter-router connection downstream router shows that qd_compose_insert_string_iterator calls are expensive.   This is due to the way qd_compose_insert_string_iterator copies data from the iterator: byte by byte.

Refactoring qd_compose_insert_string_iterator to use qd_iterator_copy and qd_compose_insert_string_n will leverage memmove semantics and will improve performance (see attached flamegraphs: 134462 current vs 38506 w/optimization)