This doesn't make sense.
The recommendation to use RSTP does refer to use of hAP ax2 or hAP ax3 as CAP. But, it doesn't refer to use of any switches or existence of L2 loops in the ethernet topology.
If there are L2 loops in ethernet topology, then RSTP would be recommended regardless whether CAPsMAN/CAP functionally is being used for wireless interfaces. Which makes it orthogonal topic.
The CAPsMAN/CAP example in the docs isn't for meshing either. Which could result in L2 loops. And, there are different protocols for that.
So, only a rogue device, or heavily misconfigured/misbehaving device, could connect to multiple access points and bridge them. Which shouldn't happen normally either.
Wireless ports are practically edge ports then. And, they should become edge ports automatically according to bridge docs
auto - same as no-discover, but will additionally detect if a bridge port is a Wireless interface with disabled bridge-mode, such interface will be automatically set as an edge port without discovery.
So, is this not working properly? Could it be causing issues, when wireless interface comes from inactive state to active - when first device connects to it - after long inactivity?
If so, then maybe setting bridge ports as edge (yes, or yes-discover) would work? I am using CAPsMAN static interface provisioning anyway - for VLAN capability on 802.11ac devices.
Though, dynamic interface provisioning probably doesn't allow to set port as edge manually, and use of RSTP is a mitigation of the some deeper issue with wireless drivers?