Data-license model for campaign + derivative outputs (inherit vs assert)

Publishable output (M2) needs an explicit, correct license on every produced dataset. `docs/output_structure.md` currently carries a generic `License (SPDX id)` placeholder on each level; this issue decides the actual model. Captured from the 2026-07-10 output-structure meeting notes with Yarik.

## The levels that need a license
- **campaign** (our aggregation, `DatasetType: study`) — draft: **CC-BY-4.0**. But is asserting CC-BY-4.0 correct for an aggregation of others' data, or should it inherit?
- **babs-study** (produced, holds raw + the derivative it made)
- **derivative** (`derivatives/-++mb1/`, processed from raw)

## Open questions
- **Inherit vs assert.** OpenNeuro raw datasets are typically CC0. Do derivatives/studies **inherit** the source dataset's license, or do we **assert** our own (CC-BY-4.0)? Sometimes it depends on the source license.
- **How to carry it.** A copied `LICENSE` file per dataset, or a **symlinked** `SYMLINK_LICENSE` pointing at the source (inheritance without duplication)?
- **CC0 subtlety.** Processed derivatives may not be independently copyrightable → CC0 might be the honest license regardless.
- **babs support.** A `babs init --license` arg would let babs stamp the chosen license at init — tracked upstream as **PennLINC/babs#379**.

## Code license (done)
mechababs **code** repo → **MIT**, added with copyright holder **Center for Open Neuroscience** (matching con/duct and the pending CON-org move, #58). Revisit the holder if #58 changes it.

## Once decided
Bake the chosen model into `docs/output_structure.md` (replace the generic `License (SPDX id)` lines) and mirror into the `OpenNeuroDerivatives/fmriprepDerivatives` opinions repo.

Relates #71 (provenance gaps), #4 (output-structure gaps), #20 (GeneratedBy); PennLINC/babs#379; #58 (org move).

Start by reading docs/output_structure.md and reviewing the listed inherit-versus-assert, license-file, and CC0 questions, along with PennLINC/babs#379 and the related issues. Done means the licensing model is decided, the generic SPDX placeholders are replaced in docs/output_structure.md, and the result is mirrored in the OpenNeuroDerivatives/fmriprepDerivatives opinions repository.

Written by the indexing model from the issue text.