Use re.search instead of re.match

Some web sites have buggy utf8 encoding or not use utf8.
We don't have to be strict, we want a wide coverage.

A bug was introduced by sorting the cache content. Resulting of destroying
the cache object and replace it with a simple dict with no saving.

- Support upgrade
- Refactorise code

Since last major update, a package failure stops the whole sync.
If a package check fails, skip it

Instead of an int vs str error

Static files are now under a directory in xdg directories (config or cache)

New paths:
$XDG_CONFIG/packages.conf
$XDG_CONFIG/sendmail.conf
$XDG_CACHE/packages.conf

Doesn't compare versions, let that to new mode

Archversion now use a new cache format which store upstream and downstream
ersions (with timestamp)

cache is updated by calling the sync command

This option allow to use a regex to exclude upstream versions before trying to
find the last one. Useful to exclude rc* version from a list without building
a complex regex_version.

Previously update mode never call upstream check function and only use cached
data.

Now it always call upstream check and only save it if -S is omitted

pacman database queries are now handled by a singleton class into pacman module.
This will avoid a huge memory consumption to the current implementation.

Allow upstream web parsing to be retryed in case of error. The number of retry
can be set with the retry parameter.

Our error exceptions are now catched globablly and __str__ method describe
the issue

Allow to check package against local pacman synced db

Add new deps to alpm

Instead of using LooseVersion from distutils (a missed deps btw), I switched to
a custom version not directly based on tuples comparaison. Because it doesn't
support different type inside tuple.

All my package are currently passing better than with LooseVersion.

We also doesn't compute the full sorted list of versions but only the max, which
is the last.

Instead of having all tools into one arch tools repository, split into
sub autonomous project.