feat: support device-local image sequence input

[hyeokjun32]

Summary

• Added image-sequence directory support for device-local Vision producer input.
• Preserved existing image/video payload behavior while recording image_sequence_file evidence for sustained runs.
• Updated README and orchestration contract docs with the new override behavior.

Validation

• PYTHONDONTWRITEBYTECODE=1 .venv/bin/python -m pytest tests/test_input_sources.py tests/test_multi_workload_sustained.py tests/test_docs_links.py -q -p no:cacheprovider
• PYTHONDONTWRITEBYTECODE=1 PYTHONPATH=src .venv/bin/python -m inferedge_orchestrator run-multi-workload-sustained --config configs/agent_multi_workload_sustained_device_local.json --output /private/tmp/.../out.json --frames 6 --vision-input /private/tmp/... --capture-process-resource-snapshot
• PYTHONDONTWRITEBYTECODE=1 .venv/bin/python -m pytest -q -p no:cacheprovider
• git diff --check

Impact

• Makes the device-local sustained starter closer to real file-backed workload validation without adding live camera or production serving scope.
• Keeps the project focused on runtime operation evidence under constrained edge conditions.

Notes

• This does not claim live camera, full YOLO service, or thermal endurance validation.
• Existing config/result contracts remain backward-compatible.