Implement signature-bundle import + trust-gating validation for federated fleet intelligence

[JSONbored]

Context

The export (#6478) and collector-client (#6479) issues produce and transport signed, anonymized calibration bundles between self-hosted instances, per #1970 (opt-in federated fleet intelligence). This issue is the receiving side: verifying a bundle's signature and applying the trust-gating/poisoning-resistance rule before folding it into local calibration or the peer-median benchmark. This issue is blocked by #6477 (the maintainer-only design spec for the trust-gating rule) — the actual poisoning-resistance mechanism must be decided before this can be implemented correctly; do not start this until #6477 is resolved and its decision is available to read.

Requirements

• Verify the HMAC signature against the key-trust scheme Design: trust-gating/anti-poisoning scoring rules for federated signal-bundle import #6477 decides before accepting any bundle contents.
• Apply Design: trust-gating/anti-poisoning scoring rules for federated signal-bundle import #6477's concrete poisoning-resistance rule exactly as specified there — do not substitute a different anomaly-detection heuristic invented in this issue.
• A bundle that fails signature verification or the trust-gating check is rejected and logged (operator-visible), never silently accepted or silently dropped without a trace.
• Opt-in only, config-as-code — mirrors Implement signed, anonymized signature-bundle export for federated fleet intelligence #6478/Opt-in collector client for federated fleet intelligence (push/pull transport) #6479's toggle.
• Fail-safe: a rejected or malformed bundle must never affect the gate's own review/merge behavior for this instance's own PRs.

Deliverables

• Bundle signature verification against Design: trust-gating/anti-poisoning scoring rules for federated signal-bundle import #6477's key-trust scheme.
• Implementation of Design: trust-gating/anti-poisoning scoring rules for federated signal-bundle import #6477's poisoning-resistance rule.
• Rejected-bundle logging/visibility for the operator.
• Unit tests: valid bundle accepted, invalid signature rejected, a bundle that fails the poisoning-resistance rule rejected, opted-out instance never processes an inbound bundle.

Test Coverage Requirements

99%+ Codecov patch coverage on the diff (src/**) — including both the signature-failure and trust-gating-failure rejection paths, not just the happy path.

Expected Outcome

An opted-in instance can safely accept a peer's anonymized bundle, verified and trust-gated per #6477's design, and folds it into local calibration without being exposable to a poisoned or forged bundle from a bad actor.

Links & Resources

• Opt-in federated fleet intelligence across self-hosted instances #1970 (parent — federated fleet intelligence)
• Design: trust-gating/anti-poisoning scoring rules for federated signal-bundle import #6477 (blocking — trust-gating/anti-poisoning design; this issue implements that design, does not redesign it)
• Implement signed, anonymized signature-bundle export for federated fleet intelligence #6478 (export — produces the bundle format this issue consumes)
• Opt-in collector client for federated fleet intelligence (push/pull transport) #6479 (collector client — transports the bundle this issue validates)

[linear-code]

JSO-4446