Need to improve the recovery logic against trailing invalid index

[piste-jp]

Currently, LTFS doesn't have a good logic to recover the 2 cases below.

1. | index | data | (incomplete) index | eod
2. | index | data | (incomplete) index eod

Case 1 could be generated by the cam backend when LTFS crashes in the middle of writing index. Case 2 could be generated by low level SCSI backends lile the sg backend at the same condition.

See detail in the PR #302.

[piste-jp]

The point is how to handle the (incomplete) index.

The spec doesn't clear this condition because it doesn't define any "incomplete" condition at all. In my sense, I want to keep the (incomplete) index for analysis when something bad things happen like

1. | index | data | (incomplete) index | eod - recover -> | index | data | (incomplete) index | index | eod
2. | index | data | (incomplete) index eod - recover -> | index | data | (incomplete) index | index | eod (add a FM before index)