release(v0.0.4): add customizable models in Dify UI and new predefined models

[lws123321]

**## Summary

This PR releases v0.0.4. It keeps the existing predefined model list and adds support for adding TokenHub models from the Dify UI, so users no longer need to wait for a new plugin release when TokenHub ships a new model.

• Customizable models (customizable-model)

  • Add customizable-model and model_credential_schema in provider/tokenhub.yaml
  • Users can configure: model name, API Key, Base URL, context size, max output tokens
  • Capability toggles: thinking mode, reasoning effort, vision, tool calling
  • Implement get_customizable_model_schema() in llm.py to build parameter rules and features from credentials

• Thinking mode / reasoning effort

  • Predefined models still use built-in allowlists (THINKING_TOGGLE_MODELS / REASONING_EFFORT_MODELS)
  • Customizable models read support_thinking / support_reasoning_effort from credentials, so new models do not require code changes

• New predefined models

  • deepseek-v4-pro-202606 (DeepSeek-V4-Pro factory direct)
  • deepseek-v4-flash-202605 (DeepSeek-V4-Flash factory direct)
  • minimax-m3 (MiniMax-M3)

• Bug fix

  • Fix select option labels Yes/No being parsed as booleans in YAML, which caused Pydantic validation failures on plugin startup (now quoted as "Yes" / "No")

• Version

  • Bump manifest.yaml and CHANGELOG.md to 0.0.4

Test plan

• Run dify plugin package and confirm packaging succeeds (version is semver 0.0.4)
• Install the plugin in Dify and confirm it starts without schema validation errors
• Verify predefined models include the three new entries and existing models still work
• Use “Add model” in the provider settings to add a custom model (e.g. a newly listed TokenHub model)
  • Toggle “Support thinking mode”, “Support reasoning effort”, etc., and confirm the parameter panel and API behavior match
• For predefined and custom models that support thinking, test thinking and reasoning_effort are sent as expected
• Smoke-test deepseek-v4-pro-202606 and deepseek-v4-flash-202605 with a simple chat and streaming response**

[crazywoola]

❌ Request Changes

Decision: Request changes

Failed Checks

Check	Status	Detail	Required action
dify_plugin version	❌ Fail	dify_plugin version must be >= 0.9.0; found 0.5.1.	Use dify_plugin >= 0.9.0 in the packaged plugin environment.

Full Check Results

Check	Status	Detail	Required action
PR content language	✅ Pass	PR title/body CJK ratio=0.0% (zh=0, en=1602, ignored_zh=0, allowed_zh<=0)	None.
Project structure	✅ Pass	All expected files present: manifest.yaml, README.md, PRIVACY.md. _assets/: yes.	None.
Manifest author	✅ Pass	author is valid.	None.
Icon validation	✅ Pass	icon exists: _assets/icon_s_en.svg	None.
Version check	✅ Pass	version 0.0.4 is available.	None.
README language	✅ Pass	README.md CJK ratio=0.0% (zh=0, en=2233, allowed_zh<=0)	None.
PRIVACY.md	✅ Pass	PRIVACY.md exists and is non-empty.	None.
Dependency install	✅ Pass	requirements installed successfully.	None.
dify_plugin version	❌ Fail	dify_plugin version must be >= 0.9.0; found 0.5.1.	Use dify_plugin >= 0.9.0 in the packaged plugin environment.
Install test	✅ Pass	plugin install test passed.	None.
Packaging test	✅ Pass	packaging check passed.	None.

Next steps

• dify_plugin version: Use dify_plugin >= 0.9.0 in the packaged plugin environment.

Please address these issues and push an update.