Don't depend on a babs fork for real data processing

## Goal

mechababs should run all *real* data through upstream `babs` — carry no fork-only changes on the path that processes actual datasets.

## Why

The babs component moves slowly, so anything we carry fork-side is high-cost to maintain and blocks publishability. Reducing the fork surface to zero (for real data) should let us simplify the rest of the mechababs work.

Landing upstream **babs #365** (combine single-app + pipeline modes) is expected to remove most of the need for fork-side changes.

## Vehicle

The *pipeline-of-1* effort (run one subject per target study with the opinionated fmriprep options) is the concrete vehicle for proving the upstream-only path end to end.

---
_Filed from hub 2026-06-08; refine framing as the pipeline-of-1 work clarifies._

Start by reviewing the pipeline-of-1 effort and upstream babs #365, since no files or tests are named in this issue. Done means real datasets run through upstream babs without fork-only changes, with the end-to-end pipeline-of-1 path demonstrating that behavior.

Written by the indexing model from the issue text.