The RW node always initialized with in_replicaset=false

Queue initialization looks like this:

local queue=require'queue'
queue.cfg({in_replicaset=true})

It seems that this queue will always be not temporary (in_replicaset=true)... but it's not true.
The require'queue call generates an internal queue_init() call, which in turn leads to switch_in_replicaset.
since we can't configure in_replicaset at the require stage, we have a switch going on.
As part of the switch, a temporary space is created and migration is started.

This happens every time on the require'queue'

In our case, this leads to the creation of a space and formatting of the schema, which blocks migration from 2.10.8 to 3.2.1 due to modification of the system space.

Trace the require'queue' initialization through queue_init() and switch_in_replicaset, then inspect how queue.cfg({in_replicaset=true}) interacts with temporary space creation and migration. Done means requiring the queue does not format or modify a system space before the configuration is applied, and migration from 2.10.8 to 3.2.1 is no longer blocked.