Add another example to inheritance principle for .json without entities

[yarikoptic]

Upon re-reading current Inheritance principle formulation, nothing seems to
forbid that, and such use in general is great since allows to generalize
common metadata across all files of that datatype.

Notes on possible side-effects from "embracing" such approach (which in
principle I think is not disallowed ATM).

• per rule 4, presence of bold.json forbids presence of another _bold.json
  (i.e with entity) on the same level. So if further specialization e.g. per
  each task- is needed, common metadata needs to be duplicated across them
  (that is what heudiconv does ATM).

  Such restrictions could potentially be elevated if we adopt
  "summarization" refactoring of inheritance principle
  Replace "inheritance" with "summarization" principle bids-2-devel#65
  since order would stop to matter and thus multiple files can apply.

• I think that bids-validators are fine as checked on a single
  ds000248/T1w.json in bids-examples and modified 7t_trt.

• I do not know if tools implement it though but since there was precedence
  for ds000248/T1w.json - they better do ;-)

[effigies]

The legacy validator required the task entity for BOLD files, but the schema validator consistently permits entity-less top-level JSON for any suffix.

[effigies]

To save other reviewers time:

link

[yarikoptic]

ping @Lestropie . Also @dorahermes - would you be interested to review as this use case might be relevant to you as well?

[Lestropie]

I don't find "Generalization of Examples 1 and 4 for a sidecar file without entities" to be an accurate description here:

• It is I believe accepted parlance---though notably currently missing from the list of definitions---that a "sidecar file" is a metadata file that possess the same stem as a data file, and only has a different extension. Which means that the proposed example describing "a sidecar file without entities" has no sidecar files.

• What distinguishes example 5 from 4 is that 5 applies if there are zero metadata fields that are unique even to the subject, let alone to each of the two runs. The contents of file "/bold.json" are applicable to all "*_bold.nii.gz" in the dataset, including all subjects.

• What's being presented as novel in each of Examples 4 and 5 are not mutually exclusive. One can have both "/bold.json" containing subject-agnostic metadata, and "/sub-01/fmri/sub-01_task-xyz_acq-test1_bold.json" containing subject-specific but run-agnostic metadata.

If the goal is to exemplify how a metadata field can be inherited from the dataset root, I would do so as an addition to an existing example rather than as a substitute alternative. Eg. Have "/bold.json" but also a sidecar for each data file. Unless software is already generating datasets that are completely absent of sidecars, and that's why you're wanting to present it as an example? If that's the case, I'm not against presenting it, but it needs more accurate description.