We should not let go of either of these, we should let go of the notion of dichotomy that infects the idea of illusion.
We tend to badly overuse the notion of illusion, and the answer to questions like this depend very heavily on what is really meant by the word illusion in particular circumstances.
You seem to have oversimplified our grasp of Self from one kind of partial reality into 'an illusion' and translated that into 'a mere convention', then jumped to the conclusion that anything contingent on a mere convention must also be a mere convention.
But you give no 1) argument against the Self being some more stable form of partial reality. And give no 2) consideration of consider whether mere conventions can have real effects.
You need to address those two things, before your complaint makes any sense. Because both of them are highly debatable.
There are many forms of partial reality
In some sense, language is always deceptive, and anything with a name is therefore an illusion. There is no rule for deciding which molecules make up any given supposed 'chair' or 'brain'. But you are still using words, so you are obviously not put off by the fact of that 'illusion'. Because we see how it is simply a simplification, and a convention for coping.
If we are going to address questions like this, we need to refine our notions of relative reality much more closely, and discard the ones that are childishly dismissive, like the notion of 'illusion' itself.
How is a collection of psychological dispositions not at thing? It has effects, it makes the world more predictable, and in order to do so, it has to exist in some sense. If that is what the Self is, then, how is that an illusion? That is not an illusion, it is an epiphenomenon: It has reality and arises from things that have reality, but we misunderstand the direction of causation (or imagine a direction in a case of mutual causation). How does that make the concept less useful, or the pattern less effective?
If reality is a continuum, instead of a collection of isolated entities, again, how does that make any of it an illusion? Each thing is an unattained approximation, but that does not, again, make the concept less useful, or the pattern less effective. Newton's use of the calculus as a basis for physics has shown us very well how unattained approximations have enough reality to base a metaphysics that works.
If all is one and distinctions are conventions, how does that make those conventions into illusions? Our language and our culture is a set of conventions, but they are negotiated conventions and hardly arbitrary. We know our culture, we hold it and use it, as a repository of earlier thinking negotiated by relative survival, the trials of politics, and a shared group aesthetic.
There is a huge distinction between "Human understanding of the world is never perfect and endlessly incomplete" and "All is illusion". The former is an obvious deduction from the modernist enterprise, the latter is a pointless canard, meant to forestall real philosophical inquiry (and protect religions from unwanted challenges.)
The notion of 'illusion' itself is as much of an illusion as anything you might try to label an illusion.
Things like 'optical illusions' are the closest meaningful use, but they are themselves useful features of sensation that do not reconcile with our overall model of space and its features. So they are gaps between two levels of multiple realization of the same emergent continuum, which have evolved quirks for different local goals. They are neither wrong nor misleading when used together in context. And when observed out of context, historically, they have guided most of us to the notion that neither the data, nor the model is paramount and uncorrupted, but both are real.
Given this range of options Free Will is unlikely to be mere convention, and more likely to be, like almost everything else, an epiphenomenon, an unattained approximation, an apparent inconsistency between multiple realizations of an emergent phenomenon, a simplification of some real effect by negotiated convention, or some other aspect of reality that does not fit well with our intuitive presumptions. But none of those things is just an illusion, each is partially illusory, but still real.
Less real things can create more real things
Traffic laws, a fiction of convenience, create delays in real time, and occasional tragedies in real lives -- you miss your court date and lose custody, you end up stuck on bridge in an earthquake and die.
Religions generate wars that obliterate millions of lives.
To the extent that personalities are epiphenomena of mental traits derived from physical systems shaped by social development, and ... Human pairing, and which actual children are born, is decided by things with a very limited claim to independent reality.
