Add original execution IDs to Temporal workflow spans

[Ehsan-hash]

Temporal resets create a new run ID, which makes related workflow traces hard to connect. Workflow spans now include temporal.original_execution_run_id when the Temporal SDK exposes it. Current temporal.run_id and run-based span identities remain separate, and older supported SDK versions omit the unavailable field.

Closes #868.

Validation on Python 3.14.7:

• Regressions fail on the original code with the missing metadata field.
• Full pinned Temporal matrix: 1.34.0, 11 passed; 1.20.0 and 1.19.0, 10 passed and one feature-specific skip each.
• A real local Temporal reset produces distinct run IDs, preserves the original execution ID on both workflow spans, and keeps separate span IDs and activity parents. This uses the SDK's standard local development server because its lightweight test server does not implement resets.
• Required make fixup, scoped Ruff and git diff --check pass. Fresh independent review covers both changed files.

The tests use synthetic local workflows and an in-memory logger. No live Braintrust account, provider model, new cassette recording, dependency or release change is involved. The full Python/provider matrix was not run locally.

AI-assisted with Codex: gpt-6.1-sol/high for integration, gpt-5.6-luna/high for implementation and gpt-5.6-sol/high for independent review.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-16T17:11:24.436883Z	5088168	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review" or "Codex (@codex) security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[Abhijeet Prasad (AbhiPrasad)]

this doesn't seem right, why do we have it?

[Ehsan-hash]

The new reset regression test uses this fixture because WorkflowEnvironment.start_time_skipping() returns UNIMPLEMENTED for ResetWorkflowExecution. start_local() provides that RPC without a hosted Temporal service.

The test performs a real reset so the original execution ID differs from the current run ID. That distinction catches accidentally recording run_id as original_execution_run_id, which an ordinary run would miss. The existing tests still use temporal_env; only this reset test starts the local server.

[Abhijeet Prasad (AbhiPrasad)]

hmm, i wonder if we can better consolidate the tests then?

[Ehsan-hash]

Consolidated the setup into the existing temporal_env fixture. Ordinary tests still use the time-skipping server; only the reset test selects the local server through indirect parametrization. Both modes now share the download cache setup, with the appropriate TTL option for each server.

The real reset assertions are unchanged. The full Temporal matrix passes: 1.34.0 has 11 passes; 1.20.0 and 1.19.0 each have 10 passes and one expected skip for the unavailable metadata field. make fixup and independent review also pass.

[Ehsan-hash]

The remaining Windows Python 3.11 shard-2 failure occurs during dependency installation for test_transformers(4.42.0): the request to https://download.pytorch.org/whl/cpu/fsspec/ times out after three retries. The same job reports successful test_temporal(1.19.0) and test_temporal(latest) sessions.

The aggregate checks-passed failure follows that shard failure. I have not changed source or tests for the external download timeout.