fix(deps): update the langchain-core dependency

[daveomri]

Bump langchain-core to support LangChain 1.x

Widens the core pin so the connector installs alongside LangChain 1.x while staying compatible with 0.3.x. Connector source is unchanged; it only uses stable APIs (BaseTool, Document, BaseRetriever, BaseLoader, ArgsSchema, ToolException).

- langchain-core = "^0.3.15"          # >=0.3.15,<0.4.0
+ langchain-core = ">=0.3.15,<2.0.0"  # 0.3.x AND 1.x

Fixes #37 (@vdusek, @MQ37): the old <0.4 pin makes pip install "langchain>=1.0.0" langchain-apify unresolvable, blocking any 1.x project.

Testing

• 410/410 unit tests pass on langchain-core 1.4.8.
• pip install "langchain>=1.0.0" langchain-apify now resolves (was ResolutionImpossible).
• All doc snippets verified green on both 0.3.x and 1.x (live Apify platform + local model).

🟡 Decision: dual-support now, defer full 1.x

This PR keeps dual support (0.3.x + 1.x) and defers the full 1.x migration to a later PR. Rationale: a full 1.x move touches the agent API and dev-deps, which would reopen the already-reviewed feat/modernize-langchain-integration work. Widening the pin unblocks #37 without that churn.

The one snag, for later: there is no agent import clean on both versions.

Option	0.3.x	1.x	Note
Keep create_react_agent (this PR)	✅	⚠️ deprecated (warns)	widest support, defers the choice
Go create_agent (future, 1.x-only)	❌	✅ recommended	matches upstream, drops 0.3.x

Docs stay on create_react_agent (works on both; only warns on 1.x). When the team decides to go fully 1.x, that becomes a focused follow-up: pin core ^1.0, bump dev-deps to ^1.0, re-lock, flip snippets to create_agent.

poetry.lock

Regenerated (poetry lock, Poetry 2.x). With current dev-deps it stays on core 0.3.80 — expected: langchain-openai ^0.3.2 and langchain-tests ^0.3.5 are dev/test-only and cap core <0.4. They do not constrain end users; 1.x runtime compat is proven by the 410 tests. One lock = one resolution, so testing both versions needs a CI matrix (follow-up).

Downstream docs (prepped + verified, land separately)

• apify/apify-docs #2683: langchain.md
• langchain-ai/docs: 7 provider .mdx pages + examples/tools_example.py

Version-agnostic (done, needed either way):

• Removed APIs (VectorstoreIndexCreator, query_with_sources) → retriever + ChatPromptTemplate chain (langchain.md, apify_dataset.mdx)
• apify.mdx: instantiate tool-list classes; call_actor uses startUrls; MCP uses client = MultiServerMCPClient(...) + transport: "streamable_http"; ApifyActorsTool prose fix
• Agent chaining imports → langchain_core.messages + create_react_agent (apify_actors.mdx, apify_web_crawler.mdx, apify_google_search.mdx)
• Model name normalized to gpt-5-mini everywhere
• No change: apify_crawl.mdx

Decision-dependent (only if full 1.x): flip agent snippets to langchain.agents.create_agent and add the langchain meta to those install lines.

Follow-ups (separate PRs, non-blocking)

• Full 1.x migration (the deferred decision above): pin core ^1.0, bump dev-deps to ^1.0, re-lock, flip agent snippets to create_agent.
• Bump apify-client to ^2.5.0, replace raw requests build-fetch with SDK default_build().
• Migrate examples/tools_example.py to the LangGraph/LangChain 1.0 agent API.
• CI matrix to exercise both 0.3.x and 1.x resolutions.

[vdusek]

One comment regarding the other deps.

And the same applies to apify-client. With this constraint setup, it won't be possible to install this package together with the new major version of the SDK.

The best practice is to use an upper-bound constraint only when you know there is an incompatibility. This applies to libraries only, of course.

[jirispilka]

Looks fine but I'll leave the review to Kuba and Vlada

[MQ37]

Thank you!

I would keep the backwards compatible create_react_agent for now and we can do breaking update in future.

[vdusek]

LGTM

Also, if you generate the changelog from the PR titles, I'd use "fix(deps): ...".