Right, this is a trap (one of many :-) in using Bayes factors - I think a good background is at Bayes Factors • bayestestR where they suggest using orthonormal coding of categorical predictors precisely for this reason. We’ll be happy to help you compute your Bayes factors, but please be advised that Bayes factors are tricky beasts. Beyond coding of factors, your choice of prior for the b class will make huge changes to the Bayes factors. Additionally, it appears your model has fitting problems as you ramped up adapt_delta and max_treedepth very high (presumably to avoid divergences/treedepth warnings) - this can introduce additional fragility in the computation of Bayes factors.
An extensive background for some problems with Bayes factors is in [2103.08744] Workflow Techniques for the Robust Use of Bayes Factors, my current best thoughts on the general topic of hypothesis testing/model selection at : Hypothesis testing, model selection, model comparison - some thoughts
Best of luck with your model!