Dangling symlinks prevents collecting unused fieldmaps

[effigies]

Unfortunately, this issue remains and had escaped my previous tests.
Even if now fmriprep do no run fieldmap estimation on fmaps not listed in b0 intents (nipreps/fmriprep#3025) , the wrangler still lists/instantiates all fieldmap objects and crash when finding broken symlinks (datalad) on line https://github.com/nipreps/sdcflows/blob/master/sdcflows/fieldmaps.py#L202 . As we run jobs in parallel on temporary clones on the HPC, we only pull the strictly required data from the session to process, and thus it crashes.

An alternative would be to relax
https://github.com/nipreps/sdcflows/blob/master/sdcflows/fieldmaps.py#L202
and let broken symlink pass if only metadata are required to instantiate the fieldmap object ( which I am not sure about).

In the reason mentioned no to apply session filters, the case where one wants to use fieldmaps from another session would still work by default if they provide no bids_filters for fmap or provide bids_filters that do not specify session, or include the other session, right?

Originally posted by @bpinsard in #359 (comment)

[effigies]

@bpinsard I think it does make sense to defer checking for file existence until we actually want to use them. Hopefully we can make validation a specific call, rather than an obligatory part of building a list of potential estimators.