Update: Harden middleware infra diagnostics and live smoke

[SI-RUI-ZHANG]

原问题

• Middleware infrastructure diagnosis was being confounded by stale repo-local gateway processes on the default dev port, so scripts/dev_gateway.py status could report healthy even when the listener belonged to an older or different worktree.
• The generic ./scripts/smoke_e2e.sh --live path had regressed into a deprecated skip while .github/workflows/live-smoke.yml still depended on it, so scheduled live smoke was no longer exercising a real gateway-to-upstream validation path.
• Source-checkout maintainer flows were still documented around installed console entrypoints even though editable console scripts can be brittle in local repo environments.

本次修复

• Added a repo-local dev runtime instance marker and exposed it through /healthz, then taught scripts/dev_gateway.py status to distinguish healthy current-runtime, healthy different-runtime, and healthy-but-identity-unknown cases.
• Restored ./scripts/smoke_e2e.sh --live as a real direct-upstream live smoke using API-key auth, preserving the existing CI workflow contract without requiring host-installed harness auth.
• Added regression coverage for the new health/runtime identity behavior and refreshed contributor/operator docs to prefer repo-local python scripts/*.py helpers for source-checkout maintainer flows.

Changed files list
CONTRIBUTING.md
QUICKSTART.md
README.md
modeio_middleware/dev_runtime.py
modeio_middleware/http_transport.py
scripts/dev_gateway.py
scripts/smoke_e2e.sh
tests/integration/test_gateway_contract.py
tests/unit/test_dev_runtime.py

Reviewer: @SI-RUI-ZHANG

[SI-RUI-ZHANG]

Superseded by #14 after rebasing onto current main to resolve the controller/docs conflicts cleanly.