I'm not an expert in this, but having said that:
The selective pressure would typically not be completely absent, but hampered enough that the game can stay on much longer than usual.
Conditions like that could be imagined to set up continued high copy numbers of virus, alongside a fierce internal competition for infection. The clone will get opportunity to diversify much more due to the long infection, so there is more of a chance for novel alleles to arise and compete against their ancestral form.
And a weakened immune system gives more leeway for incremental improvement of marginal advantages. Somewhat allegorically, a new slight advantage escaping antibodies is worth nothing in context if every extra virion who escapes falls to phagocytes and cytotoxic T-cells. But if the other components are weakened, the selection pressure is unmasked. (At least it seems to me that should be the case in theory, I'm not sure if the effect is strong enough to be very important in practice in this context).
> We know that B.1.1.529 is not a “Delta plus” variant. The figure below shows a really long line, with no previous Delta ancestors. So this likely means it mutated over time in one, likely immunocompromised, individual (see my P.S. note at the end of the post).
This really stood out to catch my attention and imagination too.
With "immunocompromised" and Sub-Saharan Africa juxtaposed, it's hard not to think HIV. If (when, really) they co-infect, SARS-CoV-2 would be working in the context of HIV reverse transcriptase, and that could potentially mean going from RNA to DNA (possibly with incorporation into host DNA) and then back again by route of regular transcription. No big stretch if that would mess with copying fidelity and introduce lots of new opportunities to pick up new sequences and corresponding abilities.