Using only implication is problematic if we add negation to chain. In
such case its meaning isn't that it implies negative value, but that it
doesn't imply. That is problem if we use use negative dependencies. New
approach uses equivalency instead of implications. We used equivalency
originally because formulas added to sat solver as dependencies of one
package would introduced limitation on package even if original package
isn't selected and that is wrong. Now instead of that we use formulas
activators. It's variable that ensures that if it isn't set than
formula is always true. Activator is package group or candidate for
which given dependency is build. This way we are able to ignore
dependency formulas that don't have to be activated because we don't
choose that package.

When we have two packages providing same package and both should be
installed, than we should handle them the same way. Until now we have
chosen first candidate that was selected and that was the one that we
used. But that is problematic if we want to install both packages.

This commit ensures that when we request a package that is provided by
more than one other package than we pass modifier to both of those
packages instead of random single one.

Until now we were adding rule that only single candidate of given
package can be installed at the time. But candidates added because of
Provides directive shouldn't do that. That is completely separate
package and we shouldn't be introducing any additional restriction.

If candidate isn't from given package group (was added because it
provides that package) than we shouldn't be checking it against version
limitation. Because what some version limitation says about some
other package we don't know yet. If we specify version we should just
ignore any candidate added using Provides.

I forgot to add it as feature. This change is impossible to detect so
it should be added to features.

Running same Script multiple times is invalid.

Probably as part of some last minute cleanup I had to remove some
variables without replacing them where they were used.

When we are not using pkgupdate from packages (for example when we are
running it in different root) then there is no reason to do replan. It
might even cause problems when we are installing clean to clean root.
This new option allows can be specified to deny replanning and that way
ensure that everything is installed at once.

Also this causes behavior that if package is requested critically and is
provided by some other package then package that provides it is critical
in plan.

I forgot to actually add it.

This reverts commit c4a74b479163b64c595fb9eb00ef3661df048a81.

New sandbox should inherit environment from parent to allow protected
access to global variables defined in parent.

When we run command we would print it twice in debug mode. I would say
once is enough and so one of those prints is now lowered to trace level.
Also remove "Arg" string in other one that was left there from previous
output format.