Make paper recovery and failure outcomes truthful

[witbrock]

Outcome

Makes paper representation recover honestly from local arXiv acquisition failures, prevents a zero-success batch from being reported as completed, and keeps aggregate workflow failures specific.

Root cause

Process-wide async proxy initialisation locks were reused across the fresh event loops created by the synchronous catalogue bridge. The represented paper workflow also treated provider identity as recovery policy and accepted a bookkeeping-level partial result even when no item succeeded. Failure-like terminal states then lost their authored aggregate reason.

Scope

• loop-neutral lifecycle for the affected internal MCP adapters
• typed, provider-neutral paper acquisition recovery
• truthful all, mixed, and zero-success batch outcomes
• backward-compatible authored aggregate terminal failure diagnostics
• seed migration and durable workflow-language documentation

Validation

• 21 affected paper workflow cases
• 81 workflow contract, identity, and durable executor cases
• 125 neighbouring opportunity/loader cases, with one unrelated pre-existing advisory-timeout expectation failing
• Python lint, JSON validation, and git diff --check

Jira: JVNAUTOSCI-2638, JVNAUTOSCI-2639, JVNAUTOSCI-2640