* removes the mutation of the new_resource.version
* adds package_class_supports_arrays for multipackage providers
* cleans up the package resource initializer and sets the package_name
correctly through the accessor method.
By mutating new_resource.version we were destroying the original intent
of what the user was requesting. This is a bug that must be fixed.
Generally, subclasses will be able to get the correct information they
need through either the version argument they were passed in
install_package(name, version) instead, or through their
`#target_version_array` method. If this breaks anything then those
providers need bugfixes as well (since this is a change to an internal
API accessible only to subclassing, and there are backwards compatible
and correct ways to get the information, this is *not* a SemVer
violating change). See the fix I made to the OpenBSD provider to
preserve the same semantics, avoid using new_resource.version where it
was inappropriate, and make the code more symmetrical where before
substantially different looking code in install_package and
remove_package had exactly the same ultimate effect.
The package_class_supports_arrays 'DSL' for writing multipackage
providers coerces the arguments to the virtual methods (e.g.
install_package) into arrays, even if the user is only requesting a
single package install. This removes most of the `is_a?(Array)`
checks from the implementation subclasses.
The cleanup of the initializer is similarly necessary so that we can
use Chef 12.5 coercions to make the package_name and version arguments
accept string, but always be Arrays. This should eliminate the rest
of the need to write `is_a?(Array)` code all over the package provider
implementations.