perf(telemetry): skip non-recording LLM trace work

[dexhunter]

Please ensure you have read the contribution guide before creating a pull request.

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

• Related: Event streaming in run_async triggers repeated Pydantic serialization #4233

Problem:

trace_call_llm builds telemetry configuration and serializes request/response content before writing span attributes, even when the supplied or current span is not recording. Those values cannot be retained by a non-recording span, so large LLM payloads incur avoidable CPU work.

Solution:

Resolve the span first and return immediately when span.is_recording() is false. Recording spans keep the existing telemetry and attribute path unchanged. The new unit test verifies that a non-recording span does not read telemetry configuration, serialize content, or write attributes.

This addresses one narrow source of the repeated serialization reported in #4233; it does not claim to eliminate every serialization path in that issue.

Performance

An instrumented helper-level benchmark used content capture with a synthetic 108,167-byte response:

	Current main	This PR
trace_call_llm, non-recording span	139.820 us	0.123 us

This avoids about 99.91% of the helper overhead in that case (approximately 1,137x). It is not an end-to-end agent or model latency claim. Recording-span behavior is covered by correctness gates and remains unchanged.

The early-guard direction was identified through autoresearch with Weco. The final patch was rebuilt on current main and hardened with recording and non-recording counterexamples.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

Results:

• pytest tests/unittests/telemetry/test_spans.py -q: 80 passed
• unit-test tox environments Python 3.10 through 3.14: all passed
• strict evaluator: recording-span contracts passed; non-recording result 0.123 us
• ruff check, pyink --check, and git diff --check: passed

The repository-wide tox run was also attempted without exclusions. Its only collection blocker was the checkout-external open_source_workspace/.agents/skills/adk-verify-snippets/scripts/verify_md.py, which is not present in this checkout. Re-running all unit tests while excluding only test_verify_snippets.py passed on Python 3.10 through 3.14.

Manual End-to-End (E2E) Tests:

Not applicable for this pure tracing guard. The unit test and strict evaluator exercise both an explicitly supplied span and the normal recording/non-recording behavior without requiring an external model request.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end. (Not applicable; see above.)
• Any dependent changes have been merged and published in downstream modules. (No dependencies.)

Additional context

The commit is cryptographically signed and verified by GitHub.