According to this post, fw-heudiconv inserts the IntendedFor field automatically. I don’t have any experience with that tool, but it looks like you set your IntendedFor field in the heuristic file. In my program, we generally use a custom function called intended_for_gen() that finds the most recent field map before each functional scan after conversion to BIDS. That way you can remove bad field maps before you assign the IntendedFor field.
I think it’s probably a mistake to assume that everyone uses the same approach for their field maps, however. While I generally default to using the closest field map before each fMRI scan, it might make more sense to grab the closest field map to each scan regardless of whether it’s before or after. I’m not sure there’s a one-size-fits-all solution to this, but those two options can probably help.
EDIT: There is also this heudiconv thread, in which @satra mentions an approach that sounds better (at least for Siemens scanners), although I don’t know if there’s a straightforward implementation out there. The method described compares the shims from different files to assign field maps to scans.