The test instantiates a sharded cluster and inserts a bunch of data, asserting the inserts work. Those assertions fail if there's an election that chooses a new primary.

It's actually valuable that extending a backup cursor is exercised with elections. Unfortunately extending a cursor is expected to fail when terms change on nodes with majority read concern off. I want my cake and to eat it too.

I'm inclined to have that test work in the the eMRC=on and off case and turn off elections (other backup restore tests are actually more likely to have elections). We can leave a TODO for once eMRC=off is removed from the server. The TODO would be to change this test to allow elections, but have the inserts be resilient to failover.