Brackup

Changes for version 1.00

RELEASE NOTE: The author/maintainer of Brackup is finally happy now, and has 40 GB of data stored on Amazon, encrypted. You can trust this now. And the file formats aren't changing (or aren't changing without being compatible with old *.brackup/Amazon formats...)
track in meta header the default (most often occuring) modes for files and directories, then don't list those for each file/dir with those mode. saves on disk space on *.brackup files
support 'noatime = 1' option on a source root, because atimes are often useless, so waste of space in metafile.
rename digestdb back to digestcache, now that it's purely a cache again.
fix memory leak in case where chunk exists on target, but local digest database was lost, and digest of chunk had to be recomputed. in that case, the raw chunk was kept in memory until the end (which it likely would never reach, accumulating GBs of RAM)
make PositionedChunk use the digest cache (which I guess was re-fleshed out in the big refactor but never used...). so iterative backups are fast again... no re-reading all files in, blowing away all caches.
clean up old, dead code in Amazon target (the old inventory db which is now an official part of the core, and in the Target base class)
retry PUTs to Amazon on failure, a few times, pausing in-between, in case it was a transient error, as seems to happen occasionally
halve number of stats when walking backup root
cleanups, strictness
don't upload meta files when in dry-run mode
update amazon target support to work again, with the new inventory database support (now separated from the old digest database)
merge in the refactoring branch, in which a lot of long-standing pet peeves in the design were rethought/redone.
make decryption --use-agent and --batch, and help out if env not set and gpg-agent probably not running
support putting .meta files besides .chunk files on the Target to enable reconstructing the digest database in the future, should it get lost. also start to flesh out per-chunk digests, which would enable backing up large databases (say, InnoDB tablespaces) where large chunks of the file never change.
new --du-stats to command to act like the du(1) command, but based on a root in brackup.conf, and skipping ignored directories. good to let you know how big a backup will be.
walk directories smarter: jump over directories early which ignore patterns show as never matching.
deal w/ encryption better: tell chunks when the backup target will need data, so it can forget cached digest/backlength ahead of time w/o errors/warnings later.
start of stats code (to give stats after a backup). not done.