Throughput harness: synthetic_referrals repeats last names every 26 items, so item 26 trips OpenEMR's duplicate check #1349

What happens

benchmark/openemr_throughput/harness.py synthetic_referrals(n) builds the last name as f"{last}{chr(65 + (i % 26))}", with last = LAST[(i // 32) % 8]. So items 0 and 26 are both "QuillfernA", items 1 and 27 are both "QuillfernB", and so on. Item 26 is created right after item 0 in the same run. OpenEMR's duplicate search then shows a possible match, the program correctly stops at "Confirm Create New Patient" (text_present 'No matches were found.' fails), and the worklist stops.

Seen on 19 October: two 64-item runs (--speed standard and --speed fast) both stopped at item 26 with nothing written. Their summaries are in the harness output, not committed. The scoreboard reran with --items 26 (#1303, Amendment 4).

Fix

Make every name in the batch unique, for example by putting i // 26 into the last name, or by using the index in the letter suffix (A, ..., Z, AA, ...). Add a test that synthetic_referrals(256) has no repeated (fname, lname) pair, and none sharing a last name with an earlier item when OpenEMR's duplicate search would match on it.