I've been doing work in the UDIM area the past few months and wanted to know how best to proceed with this item here. The approach in this patch making use of the iuser field will work, as would the proposed P1958 (albeit with a tweak to account for changes since; can be provided on request) so it would be good to bottom out on which approach is preferable. There's another fix I'm working on that would require fetching the correct filename so it will also do whatever this change ultimately requires -- either it will fill in the correct iuser or it will take advantage of the P1958 change implicitly.

Personally I would like to pass the correct image user as automatic stuff can be annoying as you need to write workarounds for other behaviors.
In C-code we normally make a copy of the image user to set the tile so we don't alter the original image user. But that isn't easy to add as you will have
an image user that isn't owned by anyone. @Campbell Barton (campbellbarton) was working on a patch related to this. We should check if it would also
solve this case.

T94051: Saving single image of image sequence: image.save_render(): Could not acquire buffer from image. / D14003: Fix T94051: Add parameters to rna_Image_save_render() would add the tile as a parameter and not use the iuser at all.
So perhaps stick what has already been decided and use indivudual parameters to override the given iuser.

image.filepath_from_params(frame=, tile=, ...)

I tend to agree that it's better to be explicit and have the caller indicate which image to use. The other fix I was working on is in the C-code so I was able to create a default/empty image user just for the portion of code required.