datahub-agent-context: langchain_registration import fails, datahub.api.entities.agent missing from the pinned acryl-datahub #18686

Describe the bug

datahub_agent_context.langchain_registration cannot be imported with the exact
acryl-datahub version that datahub-agent-context pins, so the LangChain and
Google ADK registration paths fail at import time on a clean install.

datahub-agent-context==1.6.0.16 declares:

acryl-datahub[datahub-rest]==1.6.0.6

and datahub_agent_context/_registration_core.py line 26 does:

from datahub.api.entities.agent.agent import Agent, StructuredPropertyValue
from datahub.api.entities.agent.api import API_SUBTYPE_FUNCTION, Api, ApiParam

datahub.api.entities.agent does not exist in acryl-datahub==1.6.0.6. The
package directory contains: assertion, common, corpgroup, corpuser,
datacontract, datajob, dataprocess, dataproduct, dataset, external,
forms, platformresource, structuredproperties. No agent.

I also checked acryl-datahub==1.6.0.15 and the module is absent there too, so
this does not look like a resolver picking an old version. It looks like the
registration core is ahead of any published acryl-datahub.

To Reproduce

python3.11 -m venv v
./v/bin/pip install datahub-agent-context
./v/bin/python -c "import datahub_agent_context.langchain_registration"

File ".../datahub_agent_context/_registration_core.py", line 26, in <module>
    from datahub.api.entities.agent.agent import Agent, StructuredPropertyValue
ModuleNotFoundError: No module named 'datahub.api.entities.agent'

Installing the [langchain] extra makes no difference, since the missing module
is in acryl-datahub rather than in LangChain.

Expected behavior

import datahub_agent_context.langchain_registration succeeds on a clean
install, or the pin is raised to the acryl-datahub version that first ships
datahub.api.entities.agent.

What still works

The MCP tool functions are unaffected and I have been using them without
trouble. search, get_lineage (including the column argument),
list_schema_fields and get_entities all work against a self-hosted GMS:

from datahub.sdk import DataHubClient
from datahub_agent_context import set_client
from datahub_agent_context.mcp_tools.lineage import get_lineage

set_client(DataHubClient(server="http://localhost:8080", token=None))
get_lineage(urn=..., column="public_coverage_flag", upstream=True, max_hops=3)

Only the registration decorators are blocked.

Environment

• datahub-agent-context 1.6.0.16
• acryl-datahub 1.6.0.6, installed as its pinned dependency
• Python 3.11.15, Amazon Linux 2023
• DataHub Core GMS v1.5.0.6, self-hosted