deprecate workflow acceptance self-attestation

[testikun]

Problem

Workflow's Acceptance Ledger lets the same child execute a task and declare its own criteria accepted. Runtime validates the shape and labels, but cannot verify that the evidence is true. Despite that, a rejected, missing, or malformed child ledger currently overrides the canonical agent().ok execution result.

Closes #127.

Value

Workflow success once again means a runtime-observed child execution and durable result write succeeded. Existing 0.x users can still read and run old acceptance DSL/journals/artifacts, while new output clearly identifies the ledger as deprecated model self-attestation rather than verified execution evidence.

Approach

• deprecate acceptance in OpenPI 0.5 and publish its planned removal in 1.0;
• retain the old input schema, structured child protocol, replay key, and legacy artifact reader during the compatibility window;
• annotate newly evaluated ledgers with authority: "model-self-attestation" plus versioned deprecation metadata;
• return a migration warning from agent() and label completion text as deprecated self-attestation;
• make ok depend only on child execution and result artifact persistence, never the child's acceptance verdict;
• update replay regressions so tampered rejected/missing/malformed legacy ledgers remain visible without changing canonical execution facts or suppressing successful artifacts.

Validation

• bun run check — passed.
• focused Workflow acceptance, replay, prompt, and completion tests — 41 passed, 0 failed.
• full Node 22 suite — 1,241 passed, 1 skipped, 0 failed.
• Vitest suite — 30 passed, 0 failed.

Impact

• User-visible behavior: deprecated acceptance verdicts no longer turn a successful child execution into ok:false; migration text names the distinction. No graphical or TUI component changes.
• Model-visible context/tools: Workflow documentation and result text prefer plain JSON Schema plus parent evaluation; the legacy input remains accepted in 0.x.
• Runtime/lifecycle: canonical success is separated from model self-attestation; replayed execution facts and success artifacts follow the same rule.
• Persisted config/data: no config change; new ledgers add optional authority/deprecation fields, while readers still accept legacy ledgers without them.
• Compatibility/risk: intentional 0.5 semantic deprecation; physical DSL/schema/artifact removal is deferred to 1.0.