However, it (JTAG) offers no advantage over a resident debugger.
I most certainly does: it is always there.

There's been a commentary suggesting it would be better to use JTAG, which none of the conventional MCU's in this family seem to support, though that's certainly an option with programmable logic. There's also been some commentary suggesting an ICE of one sort or another would be better. Both of these options are available. However, I prefer, for the purpose of solving my particular set of problems, to use a resident debugger. This is because (a) my code is small (<16KB), (b) a debugger can easily deal with hardware/software synchronization problems, (c) I'm not working in 'C' at this point, else I couldn't consider this approach, (d) an ICE won't deal with my programmable logic applications, and so on.
The SILabs derivatives from before they went crazy (f0xx, f1xx) are "conventional MCU's in this family" and DO support JTAG.
Whatever you "prefer" sometimes will have to yield to what is possible. I would "prefer" to drive a Ferrari, but my paycheck put me in a Chevy.
a) what does code size have to do with debugging?
b) what "synchronization" is it you are babbeling about?? You keep bringing up that issue but NEVER explain. Since all debugging takes place by stopping the processor and looking at things what is this issue you keep bringing up???
c) what does the source language have to do with it?
d) that is correct, but "the JTAG equivalent like SILabs" has been suggested and could be implemented.

BTW, we never, Never, NEVER build test hooks into deliverable code.
OOPS!! you know the NASA approach? "we fly what we test"

Erik