This test was failing every once in a while because we include the `includeArrayIndex` in unwind. This means if we have a $lookup-$unwind query, and the lookup produces results in a different order (uses a different join strategy or foreign-side access path), we'll have a different array index in the result. It's not possible to account for this difference in the final result comparison since the difference could propagate in many different ways. In this change I just remove the unwind array index field.

I used `rngSeed=1788196606693` to verify this change fixes it (I ran into this in a patch and noticed it was unrelated)