ref!: Stop disabling client integrations when an agent integration is present

[alexander-alderman-webb]

Description

Use a boolean scope member to suppress chat spans from client libraries that are inside a chat span from an agent framework. Add the traces._AgentFrameworkChatGenerationContext wrapper to ensure the scope member is cleaned up correctly when the agent framework span exits.

Remove logic that disables client library integrations when an agent library is active. Delete the integration_deactivation tox environment that ran the corresponding tests.

Use the context manager in all agent frameworks that create chat generation spans (langchain, pydantic-ai, and openai-agents). Add early returns for chat generation spans in the openai, anthropic, and google-genai client libraries to prevent duplicate generation spans.

Remove embedding patches from the langchain integration. These must now be provided by client libraries like openai, anthropic, etc.

Issues

Closes #5515

Reminders

• Please add tests to validate your changes, and lint your code using uv run ruff.
• Add GH Issue ID & Linear ID (if applicable)
• PR title should use conventional commit style (feat:, fix:, ref:, meta:)
• For external contributors: CONTRIBUTING.md, Sentry SDK development docs, Discord community

[github-actions]

Codecov Results 📊

✅ 66703 passed | ⏭️ 3050 skipped | Total: 69753 | Pass Rate: 95.63% | Execution Time: 190m 41s

📊 Comparison with Base Branch

Metric	Change
Total Tests	📉 -65699
Passed Tests	📉 -61618
Failed Tests	—
Skipped Tests	📉 -4081

All tests are passing successfully.

✅ Patch coverage is 87.76%. Project has 2262 uncovered lines.
❌ Project coverage is 89.46%. Comparing base (660a4b3) to head (a2107ac).

Files with missing lines (4)

File	Patch %	Lines
sentry_sdk/integrations/anthropic.py	50.00%	⚠️ 4 Missing and 4 partials
sentry_sdk/integrations/google_genai/init.py	50.00%	⚠️ 4 Missing and 2 partials
sentry_sdk/integrations/langchain.py	88.46%	⚠️ 3 Missing and 3 partials
sentry_sdk/integrations/openai.py	87.50%	⚠️ 1 Missing and 1 partials

Coverage diff

@@            Coverage Diff             @@
##        master       #PR       +/-##
==========================================
- Coverage    90.17%    89.46%    -0.71%
==========================================
  Files          193       180       -13
  Lines        25768     21460     -4308
  Branches      9532      7804     -1728
==========================================
+ Hits         23236     19198     -4038
- Misses        2532      2262      -270
- Partials      1432      1217      -215

---

Generated by Codecov Action

[sentrivana]

I guess this would work.

Why do we want to go via this route in the first place? What do we gain over the current way of doing things?

[sentrivana]

Why have a special property for this and not return the span directly in __enter__? Might be wrong, but from a cursory look we seem to only ever be interacting with the span when using the context manager.

[alexander-alderman-webb]

The idea would be for the agent frameworks to use this wrapper with the context manager protocol since __exit__() sets

self._span._scope._agent_framework_generation_entered = False

[alexander-alderman-webb]

Users that use both agent and client frameworks in the same process would benefit.
So, e.g., Seer does a google-genai call to search the docs inside a tool. Seer doesn't use an off-the-shelf framework but generally this seems to be a common pattern that I've seen a few times in users' traces.

[sentrivana]

OK sounds good!