Q:When to use Cortex-A vs Cortex-R vs. Cortex-M

A: I bit of clarification concerning the Cortex-R series. The "R" series is not necessarily low power, nor is it necessarily deeply embedded. Those are attributes mainly of the "M series.

The "R" series is the deterministic variants. They don't offer non-deterministic interrupt/exception servicing like the "A" series or somewhat the "M" series. The "R" series, despite being the "real-time" variant is not necessarily faster than either the "A" or "M" series. They do guarantee precise timing of interrupts and exceptions. You always know exactly how many clocks from the interrupt to the ISR. This is critical in hard real-time. One of the performance features of the "A" series (and I believe the "M: series as well) is something known as "tail chaining" of interrupt handling. It allows the processor to greatly reduce the number of clock cycles when firing nested interrupts. Unfortunately, it also makes the processing less deterministic. There are a number of other features of the "R" series I can't recall off the top of my very tired head right now.

I will agree, I've not seen any applications that use the "R" series as many people improperly confuse "real-time" with "fast". Real-time is deterministic. Being too fast is just as wrong as being too slow.