Fix the trace comparison for an async migration

An AI agent converted a 2,000 line Node module from callbacks to async/await and called it "a pure mechanical conversion, no behavior change". The tests are thin. So you replay recorded inputs through the old and new versions and compare the side effects each one produced, in order: database writes, emitted events, outbound calls.

An assistant wrote compare_effect_traces(old_trace, new_trace), which describes the first difference. It is in the editor. It has a bug. Find it and fix it.

Input and output:

• old_trace and new_trace: the effects the old and new version produced for one input, in the order they happened.
• Return one string: the first difference, or identical.

Rules the tests check:

• Compare the traces position by position, up to the length of the shorter one. At the first position i where they differ, with old effect X and new effect Y:
  • if the two traces hold exactly the same effects, each the same number of times, only in another order: reordered at <i>: expected <X> got <Y>
  • otherwise: diverged at <i>: expected <X> got <Y>
• If they match up to the shorter length and the old trace is longer: missing at <k>: <effect>, where k is the shorter length and the effect is the old trace's entry at k.
• If they match up to the shorter length and the new trace is longer: extra at <k>: <effect>, with the new trace's entry at k.
• If the traces are equal: identical.