(PUP-2692) Cache function info with the functions

The Puppet::Parser::Functions code used to hold onto two different sets
of information about the loaded functions: a hash of environment name to
environment module, and a hash of *environment* to a hash of function
name to function information. It was assumed that the
Puppet::Node::Environment class was written to properly work as a hash
key since it implemented both hash() and ==(). This assumption being
true would have meant that different instances for the same environment
would be the same key and the second hash above would not grow beyond a
reasonable size for the number of environments.

It turns out that to be a proper hash key a class must implement hash()
and eql?() (not ==()). Because Puppet::Node::Environment didn't do this
the hash of function information would just continually grow when
directory environments with an environment_timeout != unlimited was in
use. This is because the Puppet::Node::Environment instance would be
periodically thrown out by the Puppet::Environments::Cached loader and
be recreated, but the Puppet::Parser::Functions hash would hold onto the
old instance.

The fix for this here is to consolidate all of the function information
into a single structure. The function modules now have a few methods to
get the function metadata instead of it being stored separately. This is
completely backwards compatible and does not conflict with any existing
uses of the environment modules since the new methods are on the module
itself and will not be part of the mixin.