Fixes #3433: allows to continue deployment on target node var lookup error

[fanf]

Fixe for http://www.rudder-project.org/redmine/issues/3433

I'm really not sur that this fix is correct. Not the code, but the semantic of the fix, for two reason:

1/ I'm wondering if it could lead to inconsistency in list sizes.
Suppose we are in the middle of deployment, and that a node is removed in the middle of it. Can't we have one generated list of nodes (hostname for ex) of size n, and the next of size-1, and so mismatching array size in cfengine.
Actually, I'm wondering what prevent it now.

2/ Perhaps the generated promises will just be wrong. We want to apply some config depending of hostnames of nodes in a group (for a cluster, for example). If for some reason, one the hostname is not lookupable when deploying, the node will be ignored (and the config false). Isn't it better to fail the deployment in that case ?

Perhaps we are looking for the bad fixe, and what we really want is a consistency check at LDAP level, saying that you can't have node ID in groups not in ou=node (or something alike).

Fell free to refuse the pull request!

[ncharles]

I am really not convinced by this change.
Indeed, if a value that is expected is not there, it's probably better to fail rather than continue and generate promises that can be invalid.
And with the implementation here, it will skip the invalid id, and have a too short list of id that could wreck havoc if used by a variable.
I'm rejecting this