I looked into this a while back actually. I tried scanning all the original maps and seeing what cell types each tile was associated with. The results were ambiguous. Some tiles had many different cell types associated with them, and there was no clear way to automatically choose a "correct" one to set as the default. Another problem is that some tiles were never used in the original maps and so they had no cell type associated with them.
And yeah, new tile sets would need new data. Plus, the tileset and map files aren't structured to store this data, so you'd probably end up with another external file format, which is just a pain and kinda ugly. But then taking it onto say the tileset files would also be ugly at this point.
But you're right. There does need to be some way to automatically associate a tile with a cell type. I guess you could use the above method to get started, and then just have someone with too much time on their hands verify correct choices and deal with the ambiguities. The default tiles sets only have about 2000 tiles. It's a fair bit, but still doable by hand.