Sync structured research output parsing in Python SDK

[Shedna]

Description

This sync fixes a contract mismatch for the public POST /v1/research endpoint in linkup-python-sdk.

The platform supports structured research outputs, but the Python SDK returned raw dictionaries from research() / async_research() even when callers supplied a Pydantic structured_output_schema.

This PR synchronizes the SDK by:

• parsing completed structured research task outputs with the caller-provided schema in research() and async_research();
• keeping existing behavior for sourced answers and for task retrieval endpoints where only the echoed JSON schema is available;
• adding focused unit coverage for sync and async structured research responses.

Validation run:

• make lint
• make typecheck
• make test

Checklist

• I have installed prek on this project (for instance with the make install-dev command)
  before creating any commit, or I have run successfully the make lint command on my changes.
• I have run successfully the make typecheck test command on my changes.
• I have updated the README.md if my changes affected it.

[cjumel]

Closing this PR in favor of #76, where I took the opposite direction: instead of fixing the "missing" pydantic conversion (but they were not actually missing since the places in question the output is systematically None I believe), I made sure to never do this conversion for consistency, as we can't do it everywhere ourselves (e.g. when you get a task results, we don't have the pydantic model anymore)