Fix service-link worker scheduling and reconcile issue evidence

[Proxicon]

Summary

Service-link workers now persist the safe peer error and next scheduled recovery time after a verified database abort reloads the attempt. A subsequent handled peer failure used to prevent the awaited assignment from returning that replacement, leaving the worker updating its detached original object. Resolve the current tracked attempt in the existing error catch; retain pending changes, optimistic concurrency, Serializable isolation, four total attempts, cancellation and consent boundaries.

The regression uses the existing real PostgreSQL competing-write fixture: actual 40001 followed by HTTP 503 or response loss, fresh-context persistence and rollback/identity checks, then due-only recovery after restart with the original payload. Reconcile the existing beta.4 ledger and guides with merged implementation, local protected-operation receipts, exact partial native peer evidence and outstanding real-provider gates. Related to #84, #92, #93 and #128. #128's published-peer acceptance remains pending; #89/#90/#83 remain open. Dependency #122 is already merged separately as 5766c33.

Validation

• Meaningful regression added: 2 failed before correction, 2 passed after; 92 related guards passed with no failures/skips (Debug, actual PostgreSQL and synthetic HTTP contract peer).
• Release solution restore/build passed: 0 errors, 21 existing warnings; full suite 2,073 passed, 0 failed, 6 existing environment-gated skips in 29m36s, including both new actual-PostgreSQL regressions.
• Existing Slopwatch 0.4.2 analyzed both changed .NET files: 0 findings; unchanged disclosure/layout gates and whitespace passed.
• Current guides and historical evidence reconciled; relative documentation links resolve. Independent source/evidence review found no outstanding concrete finding.

Normal hosted gates validate the final intended current-main candidate. No release or upstream change is included.

Disclosure check

• No credentials, customer data, employee data, private URLs, internal hostnames or private deployment details are included. Source/image digests are public immutable artifact evidence.

[Proxicon]

Independent source/evidence review completed against head 25253da668fc226cc1049e4e817abcb4ba7a2839 and base 5766c33dce02be141d060b45d3e135eb37ee1558: no outstanding concrete findings. Both committed .NET files exactly match the reviewed correction and meaningful real-PostgreSQL regression. Resolving the retry-owned tracked attempt retains pending changes and optimistic concurrency; Serializable isolation, the four-attempt bound, cancellation, fixed deadlines and human-consent/current-authority guards remain intact.

Fresh-context error/schedule persistence, rollback, unchanged identities and principal/secret material, disabled business authority, and due-only restart recovery are covered. Results: 2 failed before correction, 2 passed after, 92 related existing tests passed (Debug); final full Release suite passed 2,073 tests with 0 failures and 6 existing environment-gated skips, including both new regressions. Slopwatch found 0 findings in the two changed .NET files. These are actual PostgreSQL/synthetic HTTP contract-peer results; real-product and publication acceptance remain separately tracked in #89/#90/#121/#124/#126/#128.

Normal hosted validation is still running. The parent will refresh the exact head, current main, test-merge tree and all actual check conclusions before using the normal merge workflow.