Study 09 data: add two human transfer cohorts

Child of #134. Draft implementation: #139.

## Stage-A outcome

- [x] Eckstein: complete 306-person public release admitted as v2; the paper reports 291 and no machine-readable exclusion list was found, so all public files are retained.
- [x] Findling: complete 22-person, 132-session analytic cohort admitted as v1.
- [x] Release identities, terms, checksums, exact inclusion audits, and source metadata are committed.
- [x] Canonical adapters preserve available subject/session/trial and task metadata.
- [x] Eckstein first-half/second-half v2 and Findling odd/even-session v1 manifests pass regression tests.
- [x] Common-Q and author-model feasibility are documented; no new author model is implemented in Stage A.
- [x] Both D=614 seed-0 GPU smokes passed.
- [x] Full GRU matrices and common-Q fits completed and were frozen with exact ordered held-out trial-key parity.
- [x] Results and representative sessions are included in the Stage-A decision report.

The data-ingestion and Stage-A execution scope is complete and in review.