Re-consider how to test diff snapshots

I mistakenly assumed that exporting a diff snapshot with a depth of 900 would result in a snapshot file with 900 tipsets. This property is not true when there are null tipsets. Exporting 900 epochs may result in fewer than 900 state-roots if some tipsets do not exist.

Are the other properties we should test instead?

Right now, we _do_ test whether the union of the diff snapshot and the base snapshot yields a complete snapshot. This test may be sufficient.

https://github.com/ChainSafe/forest/blob/8d4e42b5771270b60b5cfa50b8d8d3f0af10037a/scripts/tests/forest_cli_check.sh#L52-L54