Dear Stefan,
thank you for an interesting essay. However, a small remark on your discussion of the emergence of irreversibility: if we use the same reasoning you use to explain irreversibility in the direction of the future, we arrive---if we don't illicitly assume that the future ought to be different from the past---at the conclusion that the past ought to be likewise at high entropy.
Remember, the laws of physics are time-symmetric, so whatever argument you use to argue for a high entropy future will also result in a high entropy past; it's only by postulating a low entropy past that one gets an overall increase in entropy, a well-defined thermodynamic arrow of time, and a future-directed increase of irreversibility. Given time-reversible laws, predicting the future is the same as retrodicting the past; but to make the correct retrodiction, we have to condition on the fact that phase-space trajectories must start out in a low-entropy state.
But then, why should the past be low-entropy? In a way, this is just the original question again: why does future-directed irreversibility arise? Indeed, in the end, it's not so clear that this line of reasoning really makes any progress at all on the problem, or if the answer isn't smuggled in at the ground level.
I think there exists a similar danger in attributing intentional behavior to systems that can be modeled as intentional agents---namely, that of anthropomorphization. When a little child claims that her doll wants tea, that doesn't mean that her doll has any actual desires; this desire is an artifact of the model, not a property of the system modeled.
In the same way, basically every physical system can be described in intentional terms---like the rock you throw 'wanting to' minimize its potential energy, or generally systems wanting to evolve towards minima in the potential landscape. Clearly, that doesn't imbue the rock with any actual desires; but then, does our modeling the mosquito as wanting to suck our blood reflect a desire of the mosquito, or an artifact of the model? There certainly is a model that makes no mention of the mosquito's intention, and yet, suffices to predict its behavior, if we actually go down to the microphysical description.
Some pragmatic souls may claim that the success of intentional models over microphysical ones (which become unmanagable somewhat quickly) suffices to give them objective preference, thus arguing that yes, there is a sense in which there's some objective intentionality to the mosquito; I think that overlooks that the modeling is always done from an inescapably human perspective, which already includes intentionality. Mosquitos, ultimately, are elements of the particular way we coordinatize the world using concepts, not of the world as such; thus, we're arguing from a specific conceptual perspective---but this is itself not objective.
Anyway, I think a little more care is necessary in order to clearly avoid such pitfalls; otherwise (as you know) I think your main conclusion, that there is no intrinsic tension between goal-free microscopic and intentional macroscopic descriptions, is entirely on point.
Cheers,
Jochen