Hi Matthias,
The best word, in my opinion, to express metaphysical (but not necessarily logical) contingency is “dependent.” So, let’s say that A exists in all possible worlds. This would make A logically necessary. However, A may also be dependent on B, so that in every possible world in which B is instantiated, A is reliant upon B and exists wherever B exists.
This is one of the reasons why the TCA isn’t suspect to the “modal collapse objection” that some of the traditional Leibnizian versions of the cosmological argument run into.
Of course, a stress on “dependency” changes the argument a bit. Instead of saying that the set of all contingent entities is itself contingent (this is certainly true), we are now saying that the set of all dependent things is dependent (which I agree with, but its demonstrability requires more argumentation). We could illustrate the soundness of this claim by various analogies. If, say, the set of one type of dependent entity relies upon the set of another, and vice-versa, then the union of both sets is insufficient to explain the existence of this union.
For example, the set of all oak trees is dependent upon the set of acorns. Yet, the reverse is also true. What this shows is that if we suggest that oak trees and acorns are the only members of the set causing the union of oak trees and acorns, then we are engaging in circular reasoning, since we do not have an explanation for the existence of the union as a whole.
I think an easier way of arguing this point is not in terms of explanation, but in terms of causation qua rank or source. If the regress of dependent beings does not have a source in an independent (and arguably self-existent) being, then we have an infinite regress of dependent beings. However, given the finitude of dependent beings, it is much more likely that the regress as a whole is likewise finite (this is an inductive argument). An independent First Cause is then posited to end the regress.