Modernize Python/TensorFlow CI and add example smoke tests

[shenweichen]

Summary

This PR modernizes the project's TensorFlow compatibility, CI matrix, examples, and smoke-test coverage.

• Replace the legacy CI matrix with explicit Python/TensorFlow combinations.
• Add conditional legacy Keras support for newer TensorFlow versions via tf-keras and TF_USE_LEGACY_KERAS.
• Modernize LINE and SDNE to use public tensorflow.keras APIs.
• Remove the runtime dependency on deepctr.
• Refactor examples into importable main() entry points with lightweight smoke mode.
• Add CI coverage for examples/*.py.
• Update model tests with embedding count and dimension assertions.
• Update package metadata, dependency extras, README install guidance, and the CI badge.

Extended description

The previous CI matrix attempted broad TensorFlow coverage through version combinations that are no longer consistently installable across supported Python versions. This PR replaces that setup with an explicit compatibility matrix covering TensorFlow 1.15.5, 2.10, 2.15, and 2.20 on supported Python versions. It also updates GitHub Actions dependencies and adds version-specific install handling for NumPy, protobuf, and legacy Keras.

The TensorFlow-backed models are updated to avoid private Keras imports and compatibility-sensitive helper dependencies. LINE now uses TensorFlow-native reduction and loss helpers instead of deepctr, while SDNE now avoids NumPy operations on symbolic tensors and uses public tensorflow.keras APIs.

The example scripts are also made easier to validate. Each example now exposes a main() entry point and supports a smoke mode that uses smaller graphs, fewer epochs, and reduced embedding dimensions. This allows CI to import and run the examples without requiring full training, plotting, or evaluation runs.

Package metadata and docs are updated alongside the code changes. TensorFlow is exposed as an optional tf extra, test dependencies are grouped under a test extra, Python support is clarified as >=3.7, and the README now uses a modern editable install command.

Resolved issues

Resolved by this PR:

• 【Question】NotImplementedError: run SDNE in Python3.8, tf2.11.0 #78
• SDNE Examples unable to run #30
• SDNE #45
• SDNE,'NoneType' object is not callable #65
• 在创建图G时发生读取文件错误 #69

Related but not closed by this PR:

• Wrong：SDNE  #79: this reports a Python 3.6 / tf-gpu 1.5 / cuBLAS runtime failure, which is outside the updated support matrix.
• For incompatible dependencies issues, see my environment.yml file updated on 2024/11/19 #82: this provides an external environment reference; this PR adds project-side CI coverage and dependency extras but does not adopt that environment file directly.

Notes

• Example smoke tests use reduced graph sizes, epochs, and embedding dimensions so CI can validate behavior quickly.
• TensorFlow-dependent tests are skipped automatically when TensorFlow is not installed.
• Full compatibility validation is covered by the GitHub Actions matrix on this PR.