deps: add upper bounds to agent framework and e2e test dependencies

[ling-senpeng13]

Closes #452.

What

Adds upper bounds to 14 dependency declarations. No new files, no lockfile, nothing to regenerate later.

pyproject.toml — the agents extra, 9 lines:

Package	From	To
langchain	>=1.2.13	>=1.2.13,<2
langchain-core	>=1.2.20	>=1.2.20,<2
langchain-openai	>=1.1.11	>=1.1.11,<2
langgraph	>=1.1.3	>=1.1.3,<2
google-adk	>=1.27.1	>=1.27.1,<3
openai	>=2.28.0	>=2.28.0,<3
openai-agents	>=0.12.2	>=0.12.2,<0.13
anthropic	>=0.91.0	>=0.91.0,<0.121
claude-code-sdk	>=0.0.25	>=0.0.25,<0.1

scripts/package-e2e-bundle.sh — the heredoc that writes the bundle's requirements.txt, 5 lines: pytest>=8,<10, pytest-asyncio>=1,<2, pytest-xdist>=3,<4, pytest-rerunfailures>=16,<17, mcp-testkit>=1,<2.

The rest of the diff is comments explaining why the ceilings exist, so they don't get stripped as noise.

Why

Floor-only declarations meant conductor-python[agents] accepted any future major of its integration deps, and the bundle's run.sh resolves at run time — so a consumer got "whatever is newest today." A breaking upstream release reached users and downstream gating CI with no change here, presenting as an SDK or server regression. See #452 for the full write-up.

Not hypothetical: a langgraph/pydantic interaction with typing.TypedDict already fails test_suite11_langgraph.py::test_messages_state_detection on Python < 3.12. Only the interpreter pin keeps that green — not a dependency bound.

Verification

uv pip compile --universal --python-version 3.12 over the bounded set: 99 packages resolve, and every capped package lands on the version already installed in a known-good environment — langgraph 1.2.10, anthropic 0.120.2, openai-agents 0.12.5, claude-code-sdk 0.0.25, pytest 9.1.1, mcp-testkit 1.0.3.

So this is a no-op today. It only takes effect when someone publishes a new major.

Draft — two things to settle first

1. The 0.x policy is the real decision here. Policy applied is next-major for 1.0+ and next-minor for pre-1.0, since 0.x minors are breaking by convention. That makes anthropic>=0.91.0,<0.121 genuinely protective but tight enough that the next anthropic release needs a bump. The looser alternative is <1 for all three pre-1.0 deps — one-line change each, much less churn, but close to no protection given their cadence. I picked protection over convenience; happy to flip it. Renovate on this file would make the churn a reviewable PR rather than a chore.

2. Ceilings have a cost. They can block a consumer who legitimately needs a newer langgraph, and over-capping is a well-known source of resolution pain in Poetry libraries. <next-major is the mainstream compromise, but it is a trade worth someone else's eyes.

Considered and rejected: generating a lockfile inside package-e2e-bundle.sh. That script documents being static — "no install, no network — the pinned version does not have to be on PyPI yet, so this can run before the publish job finishes" — and compiling a lock needs both, breaking release ordering.

Out of scope, noted in #452

mcp-testkit requires Python >= 3.11 while the bundle's README.md and run.sh advertise 3.10–3.13; compiling at 3.10 is unsatisfiable. Not fixed here.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.