Ok, I've been fuzzing `sequential-storage` with unexpected shutdowns of the mock flash. It's running way better than expected.
So far I've only found bugs on my fuzzing program and once I fixed those, I've only found a crc collision.

Right now the CRC is 16-bit, so that might not be enough to be really reliable... Question is whether I'd want to extend the header size from 6 to 8 so I can use a 32-bit CRC. This would cut the collision rate from ~1/65k to ~1/4B.

If the CRC is incorrect, a stored item is counted as corrupt and skipped. This means that if you write something to flash when a shutdown (or async cancellation) happens, the item might not be retrievable, but the other items are still there just fine.

I think there might be one issue that so far hasn't been found by the fuzzer and that's the page states. But for that I have to add CRC collision detection in the fuzzer first and then let it run longer to find it.