Add CPU, memory, and accelerator resource requests to PENDING span

[morgan-wowk]

Add resource attributes to PENDING execution spans

execution.resources.cpu, execution.resources.memory, and
execution.resources.accelerators sourced from task_spec annotations.
Enables distinguishing workload types (e.g. Nebius H100 vs GKE L4)
and correlating resource requests to execution timing.

The attributes are attached exclusively to the PENDING status span, where resource reservation is most meaningful. The function try_emit_execution_trace has been renamed to emit_execution_trace to better reflect its role.

Screenshots

[morgan-wowk]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• Add CPU, memory, and accelerator resource requests to PENDING span #258 👈 (View in Graphite)
• Add pipeline/parent context to root execution span #257
• Add cache hit/miss context to root execution span #256
• Add launcher type and cluster URL to root execution span #255
• Add k8s pod/cluster attributes to PENDING status span #254
• Enrich error/success status spans with failure context #253
• Add retroactive OTel execution lifecycle tracing #252
• master

This stack of pull requests is managed by Graphite. Learn more about stacking.

[yuechao-qin]

I think it'll be more thorough to add a task_spec so that it proves PENDING is the only one that gets resource states.

        execution.task_spec = {

            "annotations": {
                kubernetes_launchers.RESOURCES_ACCELERATORS_ANNOTATION_KEY: '{"H100": 1}',
            }
        }
        execution_tracing.emit_execution_trace(execution=execution)

[morgan-wowk]

🤖 Updated — test_non_pending_spans_have_no_resource_attrs now sets a task_spec with RESOURCES_ACCELERATORS_ANNOTATION_KEY on an execution that goes straight QUEUED → SUCCEEDED (no PENDING span). Also added an assertion on execution.resources.accelerators to make the non-PENDING guarantee explicit for both resource attribute types.