feat: qualify native model swapping on AMD

[dbourdea]

Summary

• harden native model routing lifecycle, cancellation, rollback, readiness, and adjacent port allocation
• add Linux AMD process-memory telemetry and Qwen 3.5/3.6 GGUF MTP-aware geometry handling
• expand the private qualification harness and regression coverage for live two-model AMD swapping

Verification

• ruff check passes for all changed Python files
• 183 focused tests pass
• full repository run: 1947 passed, 81 skipped, 21 failed; all 21 failures reproduce unchanged at the parent commit on the same AMD host
• live LAN-215 two-model qualification passed with restoration, direct/warm/cold/A-B-A routing, concurrency, cancellation, failed-switch rollback, restart re-adoption, TTL, metrics, logs, auth, and cleanup

Scope and safety

• no llama-swap or llama.cpp source changes
• raw host artifacts remain private
• service deployment is handled separately from this source review