Would the H really appear in this case though?
Yes, it would. The meaning of flag "H" is exactly to show when operations on certain port are offloaded to hardware. If state of flag is not in sync with actual operation, it's a bug. But I don't remember a report when HW offload was active, but "H" flag was not shown.
And, your observations on hEX and hAP ac2 are expected (as are your assumptions).
Also after setting the bonding interface on my home hEX exactly as on the office RB5009 and adding it to the bridge, the H does not appear on that interface
That's because hEX doesn't support bonding in HW (read
this article, it has a part explaining HW offload support). It seems that RB5009 is in the same boat (for now). So yes, if you use bonding on RB5009, that means all traffic between bond and the rest of ports will have to pass switch-CPU interconnect and will have to be dealt with by CPU. Since RB5009 has a pretty powerful CPU, it might even support throughputs higher than speed of individual bond link (i.e. more than 1Gbps), but devices with slower CPUs (such as hEX) would struggle (if not choke the switch-CPU interconnect) and there's no sense in having bonds other than for redundancy reasons (for which something based on xSTP might make more sense performance wise).