Small performance enhancement for parsing result sets (text and binary)

Two per-row hot paths will be streamlined, with no functional change:

• Binary rows (mthd_stmt_fetch_to_bind(), used by mysql_stmt_fetch()): loop-invariant statement attributes will be read once per row instead of once per column, the temporal length check and the CONC-820 limit checks will be merged into a single switch on the column type, bind row pointers will only be reset for NULL columns or on error, and conversion functions and result callbacks will advance a private copy of the position so the next column starts at the already validated column boundary.
• Text rows (mthd_my_read_rows(), used by mysql_store_result()): the row node and the row data will be allocated in a single memory root allocation instead of two, and a destination bounds check that is implied by the packet bounds check will be dropped.

Fetching lots of rows result about 8% less client CPU in both cases.