feat: make pandas an optional dependency

从 pyproject.toml 开始，检查 client.dataframe.* 和 client.records.create(table, dataframe) 中对 pandas 的引用。验证核心安装不再需要 pandas，同时 dataframe extra 会恢复现有集成；不安装 extra 时导入 client.dataframe 应产生所提议的明确 ImportError。

Summary

pandas is a required dependency (pandas>=2.0.0 in pyproject.toml) but is only used by the DataFrame integration (client.dataframe.*, client.records.create(table, dataframe)). The core SDK surface (client.tables.*, client.records.* with dict input, client.query.*) does not import pandas at all.

This makes the SDK unsuitable as a lightweight dependency for services that need only schema manipulation (e.g., table creation, column addition, relationship management) without DataFrame support. pandas adds ~30MB to the installed footprint.

Proposal

Move pandas to an optional extra:

` oml
[project]
dependencies = [
"azure-identity>=1.17.0",
"azure-core>=1.30.2",
"requests>=2.32.0",
]

[project.optional-dependencies]
dataframe = ["pandas>=2.0.0"]
`

Users who need DataFrame integration install pip install powerplatform-dataverse[dataframe]. The core SDK installs with zero heavy dependencies beyond azure-identity and requests.

Motivation

The Cowork (M365 Copilot) runtime is evaluating the SDK as the backend for Dataverse schema operations in the App Builder experience. The runtime already has azure-identity, azure-core, and requests -- but adding pandas (~30MB) as a transitive dependency for unused DataFrame features is a container footprint concern. Making pandas optional would unblock SDK adoption for schema-only use cases.

Backward compatibility

No breaking change: pip install powerplatform-dataverse[dataframe] gives the same behavior as today. Users who import client.dataframe without the extra get an ImportError with a clear message suggesting the extra.