Merge lp://staging/~invidian/intltool/fix-dangling-lock-file into lp://staging/intltool

I tested it locally on some builds and it works fine, but I can't confirm that it does not introduce a regression in race for which lock was originally added.

In this case, it is probably OK to remove the lock file, since it is a one-time race.
After the first cache-writer is done, any amount of cache-readers can use the fully created caches.

The original problem was that cache-readers started to use partial caches before the cache-writer was done writing.

> In this case, it is probably OK to remove the lock file, since it is a one-
> time race.
> After the first cache-writer is done, any amount of cache-readers can use the
> fully created caches.
>
> The original problem was that cache-readers started to use partial caches
> before the cache-writer was done writing.

I have to disagree here. flock works at the inode level, so when you delete the lock file, the next process to acquire the lock will open a new file and try to lock a different inode. We did this exact thing at Endless and it immediately broke. You really want the lock file to live as long as possible.

I think the correct fix is just to update the Makefile.in.in so that it cleans the lock file. Then you know at least all the processes make launched are done.