ci.yml: pin ci-result's needs, decouple breg-tutorial from Evidence changes, inventory the composition job, raise the breg-contracts budget

1. Nothing pins `ci-result`'s needs list. Conditional jobs such as `rust-tests`, the four `*-contracts` jobs, `identifiers` and `docs-archives` gate only through `rust-result` or `docs-required`, and `docs-required` itself is absent from `ci-result`'s needs. A test in `.github/scripts` (or `release/scripts/check-gates-inventory.py`) should hold the list.
2. `registry-mint` links `registry-evidence`, so an Evidence change selects `breg-tutorial` through the mint reverse dependency. The classifier should say whether that is intended.
3. `check-gates-inventory.py` carries no entry for the `breg-evidence-composition` job; removing the job would not fail the inventory.
4. `breg-contracts` has `timeout-minutes: 20` and runs within a minute of it. Decision pending: raise the budget (30 would match `breg-evidence-composition`) or shorten the journeys step, which is the largest share.

| Run | Duration |
|---|---|
| main, typical | 17 min 45 s |
| PR #865 heads, typical | 19 min 06 s to 19 min 29 s |
| PR #865 fifth head, attempt 1 | cancelled by the job timeout at 20 min 11 s |
| PR #865 fifth head, rerun | 16 min 49 s |

Origin: review of PR #865.

Start with ci.yml and release/scripts/check-gates-inventory.py, then inspect the tests in .github/scripts that cover gate dependencies and inventory entries. Confirm the intended registry-evidence/breg-tutorial classification and the breg-contracts timeout decision with maintainers; done means the gate list, composition-job inventory, dependency behavior, and budget are represented and tested.

Written by the indexing model from the issue text.