"we know that the failure region (Inverse Reserve Factors greater than 1) is not gonna go away at all even if that's a singularity"
Just because there is a singularity in the model, does not mean the failure region cannot be made to go away. For the element size you have, if you change the design to reduce the stress, such as the two suggestions I gave above (rotate layup 90 degrees, double the layup), the failure region may very well disappear.
I understand your point that the failure may reappear if you make the elements small enough.
That is where experimental validation comes to the rescue. A company using composites could make composite single or double lap shear coupons to measure the strength of adhesively bonded joints planned for the design. An FEA model of that coupon can be built. A specific element size can be used in that model, and model parameters found so the model results correlate to the experimental results. You can see examples of this in the literature.
You have to do the work to eliminate the failure. That may require a far more detailed model that includes the adhesive. There are designs that require a fillet of adhesive to eliminate the singularity such as I have shown in your other thread on singularities.
Below is an image of how a thickness discontinuity can induce a delamination failure mode in a composite panel.
Below is a design of the plate and adhesive profile that reduces the stress at the discontinuity.
These figures are from the paper, Techniques to reduce the peel stresses in adhesive joints with composites by L.F.M. da Silva and R. D. Adams.
Sangwook Lee wrote his PhD thesis on Failure of Laminated Composites at Thickness Discontinuities...
The samples in that work were flat panels with a stiffener in the center of the plate, but the stiffener was co-cured with the panel, so there was no additional layer of epoxy adhesive.
That thesis has experimental measurements of the onset of failure in samples, and a FEM to analyze stress in each ply. In the section on the Finite Element Model, he mentioned that his model has a singularity and why that was acceptable.