I believe I have found a solution yet I do not understand exactly what is going on. I will have to investigate a little bit further.
As it turns out, I only needed to increase h.samplerate from 858 to 889 and adjust backporch by 20 to recenter the image. This gives me a 720×480 that is displayed at correct 4:3 ratio with pillar box. As far as I can tell, these settings produce the exact same result as if I had setted my display in 4:3. I guess could measure everything to see if h.samplerate needs more adjustment because I could not predict or derive that 889 would be the proper value. To me, at this point, it seems very random but I am sure there is some logic behind this that I do not understand yet.
On my display, 720×480 gets multiplied by 4 (4k tv) in all directions without getting stretched to fill all vertical resolution. I believe I could try a 640×480 using the same method, only increasing samplerate and adjusting the backporch which my display should handle properly and apply stretching to fill the whole vertical space. I will post more info tomorrow about my findings.
If someone might enlighten me on why increasing only h.samplerate produced the appropriate AR, I would greatly appreciate it. I am sure it could have been predicted or calculated and that these values are not random at all. I will try to predict proper values before trying the the 640×480 experiment tomorrow out of curiosity but if I cannot figure it out, I will try incrementing the values one unit at a time until I get a result (or not) like I did this time.
Again, thanks everyone and thank you @marqs for creating the OSSC, it trully is a wonderful device!