fix(cluster-manager): announce only persisted endorsement changes

[RecursiveIntell]

Summary

• Emit truststore change notifications only after a newly received endorsement has been successfully persisted.
• Keep duplicate endorsements, identical re-pins, empty merges, and other tested no-op operations silent.
• Keep failed endorsement writes silent and leave the in-memory truststore unchanged.
• Add focused regression coverage for direct endorsement merges, identical-pin endorsement merges, duplicate suppression, persistence failures, missing peers, and concurrent submissions.
• Bump nvpair-cluster-manager from 1.1.4 to 1.1.5.

This fork-local PR carries the truststore notification correction and its follow-up regression coverage into RecursiveIntell/Personal-AI-Router. It is intentionally separate from NVIDIA PR NVIDIA#62; no action on the NVIDIA PR is part of this publication.

The change is based on the existing PAIR truststore implementation in this fork and preserves the repository’s existing NVIDIA copyright and Apache-2.0 licensing notices. The correction commits and follow-up tests are authored by Josh Stevenson.

Scope

Changed paths are limited to:

• services/nvpair-cluster-manager/truststore.go
• services/nvpair-cluster-manager/truststore_announce_test.go
• services/versions.json

The change does not claim to strengthen cryptographic trust, pairing-PIN security, authorization, certificate validation, endorsement verification, or mTLS configuration.

Validation

Validated candidate head:

da7df0a48060d7e729519fb5d6884778fb604c29

Base:

13b68115fa2c9c1d94f1ead1358f8d5a527cfecf

The candidate passed the cluster-manager normal and race suites, go vet, formatting, diff checks, and JSON validation in the review environment. Retained validation records report additional repository and integration checks. Repository-wide make test remains non-green in retained records, involving VCS-stamping and LM Studio flush-deadline failures also reported on clean main; those failures are not attributed to this three-path correction.

Separate private-fleet evidence reports the behavior patch active in a tested 1.1.5 runtime. That runtime evidence is bound to source commit 13b68115fa2c9c1d94f1ead1358f8d5a527cfecf. Candidate head da7df0a includes the behavior correction, follow-up tests, and version metadata; the exact candidate head is not presented as the newly activated runtime.

Publication boundary

This PR targets only:

https://github.com/RecursiveIntell/Personal-AI-Router.git
refs/heads/main

No NVIDIA upstream branch, PR, review, comment, or merge is modified by this operation.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	⚠️ Failed 2026-09-15T10:51:14.995879Z	da7df0a	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.