[issue289] mq should create & maintain diffstats in patches

[Jeremy Fitzhardinge]

Chris Mason wrote:
> Thanks for benchmarking.  As long as the time to apply (hot cache) doesn't 
> change, I'm happy ;)
>   
I think the difference should be pretty minimal.

I looked at unifying the hg import and mq patch parsers.  It would 
certainly be possible, but I'm not sure if its worthwhile.  The mq one 
is a lot more elaborate,   in an attempt to be as DWIM as possible when 
dealing with patchtools/quilt/randomness patches. 

Should hg import be as lenient?  There's already a difference in that hg 
import won't touch an RFC822-looking file unless forced, because it 
can't really parse email headers.  Perhaps the fix is to make it 
properly parse email headers (it might be useful for directly feeding 
email into mq).