Normalize workflow naming across runtime and authoring contracts - #776

Workflow authoring used overlapping names for definition metadata, run metadata, action results, and participant deliveries. This change establishes one naming contract across the runtime, CLI, schemas, bundled examples, documentation, and shipped skills.

• Separate context.workflow, context.run, context.initiator, and context.action; use explicit role and action metadata fields.
• Use expect.delivery, expect.verdicts, deliveries, and workflow deliver for participant submissions.
• Use verb-first action IDs, builtin:collect-worktree-context, actions..output_path, backend.inherit_env, and iteration_limit_reached.
• Reject retired names, including optional expression references, without compatibility aliases. Remove the unused legacy template renderer.
• Add a repository naming check with positive custom-data fixtures and negative protocol fixtures.

This is the naming slice before D3. The collector still collects one Git directory; handoff migration and additional collectors remain separate work.

Validation: make check (152 script tests), CLI build and smoke checks, 291 CLI unit tests, 112 CLI integration tests, 3147 app tests, and make build-app passed. Four sequential adversarial review rounds cover runtime boundaries, schemas, references, and the naming check.

Fourth-round follow-up fixes nested YAML custom-data scope and sequence-key ordering, with regression tests that preserve detection of genuine sibling declarations. Flow-style YAML scanning remains a documented optional coverage gap; runtime parsers still reject the retired keys. App build and make check were rerun for this script-only follow-up.