Opening car blockstore with different header damages car

[willscott]

See f6ad38b#diff-4c1ca4e17f4dada6bf68511fdf7e2cd6a0cfad180cac9bd3afa4074d3021d78bR62

With the return after the creation of a car with a proxy root, a valid car is created, and car verify complains
header lists root(s) not present as a block:

When the fix-up is attempted, the re-open fails, saying:

reopening for final root 36 vs 36
2021/09/29 11:25:42 cannot resume on file with mismatching data header

(Note: the new root being requested is the same length as the initially used one).

This is fine on a strict check, I suppose. What isn't okay is that in this case the car file is corrupted. attempting to list it reports:

2021/09/29 11:33:25 invalid data payload offset: 0

The car file should not be edited by a failed open command.