Allow reconciliation if single dataset is missing snapshots

[tschettervictor]

#53

[tschettervictor]

If you do like this, you'll have to adjust the tests.

[aaronhurt]

Overall, I'm good with this if we can make existing tests pass - meaning it doesn't alter existing expectations and is a non-breaking change. It may be time to move snapshot validation out into it's own function.

[tschettervictor]

@aaronhurt

I've merged the loops.

This is working with the following tests

Run the script a few times.
Create a snapshot of a dataset and zfs send/recv it to another nested location.
Delete the snapshot you just created.
Run with ALLOW_RECONCILIATION=0 and make sure it skips.
Run with ALLOW_RECONCILIATION=1 and make sure it completes.

I had to change the loop to loop through the source snaps as opposed to the destination snaps. This is because the source is the authoritative side of the set.

[tschettervictor]

I'm now soring snaps in reverse as the -r will remove them recursively, causing an error when the deeper one are trying to be destroyed.

I tried to see what you were doing with the tests, but can't seem to figure it out quite.

[tschettervictor]

Been running this now over a couple days, doing cloning, send/recv etc...

Works great.