According to the logs of a failure, a refresh from a stale CSRS node did not allow the shard to acknowledge 2 splits. As a result, the shard auto-splitted more than needed.

Make this check more loose, allowing the presence of an additional chunk.

The failure has only been observed on v4.4 branch, probably because in v5.0 the commit of the chunk split happens with majority, while in v4.4 the default write concern was 1.