need automatic recovery from truncated pack/index files

Note that I think fsyncdata may be needed, on the files and the
directories. (file content and metadata being separate..)

Martin Pool пишет:
> - teach check to hash the files and move away any that aren't consistent; perhaps give check an option to do only this quick/cheap check

Please, please, please.

One man from hg community all the time telling people how slow `bzr
check` comparing to `hg verify` and how it's important for him.
I think, given in mind that we have less pack files in post pack-0.92
formats than "knit" files in older formats and hg, we can do much faster
check by simply hashing files without analyzing its content. But I have
a question though: can index files be hashed efficiently? Looking at the
index file header I don't see there any checksum info.