fix(runtime): offload async inference work

[rylinjames]

Summary

• run non-batched predict_async() calls in the default executor instead of blocking the event loop
• run _predict_batch_sync() from the batch worker through the default executor as well
• add a generic ORT I/O Binding path for the denoise loop so constant inputs are bound once per chunk and per-step dynamic inputs use run_with_iobinding() when enabled
• add regression tests proving both async paths yield while slow sync inference is running, plus a fake-session I/O Binding test that avoids session.run()
• clean up stale server/test imports and add the missing os import used by the curate uploader path

Tests

• /Users/romirjain/Desktop/building\ projects/fastcrest/tether/.venv/bin/ruff check src/tether/runtime/server.py tests/test_server.py
• PYTHONPATH=$PWD/src /Users/romirjain/Desktop/building\ projects/fastcrest/tether/.venv/bin/python -m py_compile src/tether/runtime/server.py tests/test_server.py
• PYTHONPATH=$PWD/src /Users/romirjain/Desktop/building\ projects/fastcrest/tether/.venv/bin/python -m pytest tests/test_server.py::TestTetherServerWithMockORT::test_predict_async_offloads_non_batched_predict tests/test_server.py::TestTetherServerWithMockORT::test_batch_worker_offloads_sync_batch_predict tests/test_server.py::TestTetherServerWithMockORT::test_denoise_uses_iobinding_when_enabled tests/test_serve_e2e.py -p no:cacheprovider

Note: tests/test_chunk_budget_integration.py could not run in this local venv because onnxruntime is not installed.

🤖 Generated with Claude Code