The exceptions.py file existed in both the root directory and langextract/ directory with identical content. This removes the duplicate from the root to avoid confusion and maintain proper package structure.

…le (#97)

Introduces a provider registry system enabling third-party providers to be dynamically registered and discovered through a plugin architecture. Users can now integrate custom LLM backends (Azure OpenAI, AWS Bedrock, custom inference servers) without modifying core LangExtract code.

Fixes #80, #67, #54, #49, #48, #53

Key Changes:

**Provider Registry** (`langextract/providers/registry.py`)
- Pattern-based registration with priority resolution
- Automatic discovery via Python entry points
- Lazy loading for performance

**Factory Enhancements** (`langextract/factory.py`)
- `ModelConfig` dataclass for structured configuration
- Explicit provider selection when patterns overlap
- Full backward compatibility maintained

**Plugin Example** (`examples/custom_provider_plugin/`)
- Complete working example with entry point configuration
- Shows how to create custom providers for any backend

**Documentation**
- Comprehensive provider system README with architecture diagrams
- Step-by-step plugin creation guide

**Dependencies**
- Move openai to optional dependencies
- Update tox.ini to include openai in test environments

**Lint Fixes**
- Add appropriate pylint suppressions for legitimate patterns
- Fix unused variable warnings in tests
- Address import and global statement warnings

No anticipated breakage - full backward compatibility maintained. Given significant internal changes to provider loading, issues should be reported if unexpected behavior is encountered.

Add common development files, tools, and temporary file patterns

- Show current approach using factory.create_model()
- Add note that direct model passing to extract() is coming soon
- Keep planned API as commented code for reference

Ensure providers are loaded before pattern matching to prevent API key
errors when using local models. Optimize to skip loading when provider
is explicitly specified.

- Add proper permissions (issues: write for comments)
- Skip draft PRs to avoid noise
- Prevent duplicate comments with hidden marker
- Search both title and body for issue links
- Support all keyword variants and cross-repo references
- Count unique users for reactions, not total count
- Include 'write' permission for maintainer override
- Add concurrency control for rapid edits
- Handle cross-repo issues gracefully

- 6 tests: plugin discovery, loading, idempotency, error handling
- Smart CI triggers for integration test on provider changes
- New tox environments: plugin-smoke and plugin-integration

* Adding model and config parameters to extract()

* Adding extract precedence tests

* Applying changes from #120

Enable providers to define custom schema implementations via BaseSchema abstraction.
Add property-based fence output, FormatModeSchema for JSON/YAML providers, and
move GeminiSchema to providers/schemas/.

- Add HF-style patterns (e.g., meta-llama/Llama-3.2-1B-Instruct) to Ollama registry
- Add gpt-oss pattern to support issue #116
- Add comprehensive documentation for using Ollama with extract()
- Include example for direct provider instantiation when ID conflicts exist
- Add test coverage for HF-style model ID patterns

Test that format='json' is correctly passed to Ollama API

Show both ModelConfig and direct model_id usage patterns