We now use PipeFile to handle complex file type and compute md5 and filesize.

This commit add the following features in installsystems:
- A download progress bad
- check size before download
- check size after download
- check md5 after download

$ is del description
Error: not a gzip file

$ is del description
Error: Unable to open image description: not a gzip file

This is a better way of doing things. This help to have a better display of
where is causing latency

This fix is a workaround taking a class gzipstream from redhat (GPL2)
which allow gzip class to be not seekable.
Since python 3.2, gzip module is now not seekable. Until we swith to version
we should use this hack

files can be a list and/or unix filename pattern

this cause block device to be detected as an invalid destination during payload extraction

Arrow is now relative this allow to pass context of arrow
no more verbose mode. Soft is verbose except in quiet mode

Script is now compiled before execution to have a better debugging features.
local directory is no more given for each scripts
global directory is new to each script execution
only namespace object must be used to share data across scripts

Concept of data was changed to payload.
Payload is not necessary a tarball of a file or a directory, but, directly a file without tarball around.
This is useful have the same md5 on data/payload and win some space during repo adding.
Now data which come with image are represented by class Payload and all operation are located inside. This is cleaner!
Payload also handle orginial information (mtime,uid,gid,md5,size,mode)

This changement require a lot of changement in Repository and RepositoryManager class