feat: add no-op NLP engine

[ultramancode]

Change Description

Adds a NoOpNlpEngine for analyzer configurations where the active recognizers do not need artifacts produced by an NLP engine.

This is mainly useful for recognizers such as HuggingFaceNerRecognizer, which runs model inference directly and does not need NLP artifacts from spaCy. In that setup, loading a spaCy model only adds startup cost.

With this change, users can select nlp_engine_name: no_op in analyzer configuration instead of configuring a real NLP model that will not be used.

Includes regression tests and an updated HuggingFaceNerRecognizer example.

Issue reference

Refs #2012

Checklist

• I have reviewed the contribution guidelines
• I have signed the CLA (if required)
• My code includes unit tests
• All unit tests and lint checks pass locally
• My PR contains documentation updates / additions if required

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds a NoOpNlpEngine option to Presidio Analyzer to support running analyzers/recognizers without loading NLP models/artifacts (returning empty artifacts instead), and integrates it into engine/recognizer providers and model installation.

Changes:

• Introduces NoOpNlpEngine and a default YAML config for it (conf/no_op.yaml).
• Updates NlpEngineProvider, model installation, and recognizer registry/provider to support/guard no-op behavior.
• Adds comprehensive tests for initialization, batch/text processing, and Analyzer/provider integration.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
presidio-analyzer/presidio_analyzer/nlp_engine/no_op_nlp_engine.py	Implements the new no-op NLP engine returning empty artifacts.
presidio-analyzer/presidio_analyzer/nlp_engine/nlp_engine_provider.py	Registers no_op as an available engine and constructs it from config.
presidio-analyzer/presidio_analyzer/nlp_engine/init.py	Exposes NoOpNlpEngine in the package exports.
presidio-analyzer/presidio_analyzer/conf/no_op.yaml	Adds a config preset for the no-op engine.
presidio-analyzer/install_nlp_models.py	Skips model installation for no_op.
presidio-analyzer/presidio_analyzer/recognizer_registry/recognizer_registry.py	Skips NLP recognizer registration and blocks NLP recognizer retrieval for NoOpNlpEngine.
presidio-analyzer/presidio_analyzer/recognizer_registry/recognizer_registry_provider.py	Prevents configuring NLP recognizers when using NoOpNlpEngine.
presidio-analyzer/presidio_analyzer/predefined_recognizers/ner/huggingface_ner_recognizer.py	Updates docs example to use no-op engine for recognizer-only flows.
presidio-analyzer/tests/conftest.py	Adds no_op to the parametrized NLP engines used in tests.
presidio-analyzer/tests/test_no_op_nlp_engine.py	Adds unit/integration tests for the no-op engine and its provider/analyzer behavior.

[ultramancode]

NlpArtifacts does not accept a keywords constructor argument. Keywords are derived internally from lemmas in NlpArtifacts.__init__, so passing lemmas=[] keeps artifacts.keywords as an empty list.

[omri374]

Thanks! Have you seen the slim nlp engine? https://github.com/microsoft/presidio/blob/main/presidio-analyzer/presidio_analyzer/nlp_engine/slim_spacy_nlp_engine.py

Can we use this as is / adapt it?

[ultramancode]

Thanks for the pointer! I looked at SlimSpacyNlpEngine, and I agree it is relevant here.

The main difference is that slim is still spaCy-backed: it disables NER, but still initializes a spaCy pipeline for token/lemma artifacts. The use case here is a bit stricter: recognizers that do not need those artifacts at all, so we can skip initializing a spaCy model or blank pipeline.

We could technically adapt slim with a no-op mode, but that would broaden its contract: depending on configuration, slim would either provide token/lemma artifacts or return empty artifacts.

I went with a separate no_op engine to make the path clearer for cases where the recognizers themselves don’t need NLP artifacts, but I can adapt the PR if you’d prefer to keep that under slim.