Preserve GPU stream dependencies in PyTorch conversion

[XMaroRadoX]

GPU kernels currently depend on their CPU launchers but lack same-stream predecessors, allowing simulators to issue them too early (#215).

Preserve Kineto device IDs through linking and chain GPU operations by timestamp within each device/stream, retaining input order for ties and existing launcher edges. Use linked timestamps for explicit synchronization checks instead of unset protobuf start times. Skip stream chaining when device, stream, or timestamp metadata is missing; older linked traces need relinking.

Validation on a SLURM compute node: 143 tests passed. A reconstructed converter case using the four kernel records attached to #215 fails before this change and passes afterward. The attachments are JSON fragments, so this is not a complete linker reproduction. Regression tests cover collectives, copies, separate devices/streams, ordering, missing metadata, synchronization, duplicate edges, and cycles.

No added Ruff diagnostics; E501 and SIM103 on unchanged lines also occur at the base commit. Test formatting and git diff --check pass.

Refs #215.

[github-actions]

MLCommons CLA bot:
Thank you very much for your submission; we really appreciate it. Before we can accept your contribution,
we ask that you sign the MLCommons CLA (Apache 2). Please submit your GitHub ID to our onboarding form to initiate
authorization. If you are from a MLCommons member organization, we will request that you be added to the CLA.
If you are not from a member organization, we will email you a CLA to sign. For any questions, please contact
support@mlcommons.org.
0 out of 1 committers have signed the MLCommons CLA.
❌ @XMaroRadoX
_{You can retrigger this bot by commenting recheck in this Pull Request}

[XMaroRadoX]

recheck

[XMaroRadoX]

recheck

[XMaroRadoX]

CLA onboarding is complete for @XMaroRadoX. I retriggered the bot with an exact recheck; workflow 35798860698 ran the CLA action and still reports the contributor as unsigned. Could a maintainer verify that this GitHub ID has synced to the MLCommons CLA registry? The Python tests, Python lint, feeder tests, end-to-end tests, CodeQL, and C++ lint runs also remain action_required and need maintainer approval.

[marwanahmed207]

recheck