fix(analyzer): support per-entity context words (#1711)

[developer-rpai]

Change Description

Adds per-entity-type context words, resolving the use case in #1711: a recognizer that supports multiple entities (e.g. one TransformersRecognizer for ACCOUNTNUMBER, CREDIT_CARD, IBAN_CODE, US_ITIN) can now scope its context words per entity type, so "tax id" no longer boosts ACCOUNTNUMBER detections and "account" no longer boosts US_ITIN detections.

• EntityRecognizer.context accepts a dict mapping each entity type to its own list of words, e.g. context={"ACCOUNTNUMBER": ["account"], "US_ITIN": ["tax", "itin"]}. Malformed configs fail fast in the constructor (TypeError/ValueError).
• LemmaContextAwareEnhancer selects the words applicable to each result's entity type (_get_context_words_for_entity); entity types with no entry get no boost.
• Recognizer YAML configs (BaseRecognizerConfig, CustomRecognizerConfig, per-language LanguageContextConfig) accept the same dict form for context, widened from List[str].
• PatternRecognizer context goes through the same constructor validation.
• Docs: docs/analyzer/index.md documents the new form.

Issue reference

Fixes #1711

Behavior changes

• New, opt-in capability only. The flat-list form of context is byte-for-byte unchanged in behavior: the enhancer still boosts when a surrounding word matches any word in the list.
• With a dict context, an entity type with no entry receives no context boost (previously impossible to express).
• EntityRecognizer(context=...) and PatternRecognizer(context=...) now validate the value at construction and raise TypeError/ValueError on malformed input (non-string words, non-dict/list values) instead of failing silently at analysis time.

Test evidence

New presidio-analyzer/tests/test_per_entity_context_words.py (8 tests):

• unit tests for per-entity word selection (dict / flat list / None) and constructor validation errors;
• integration tests through LemmaContextAwareEnhancer + spaCy en_core_web_lg: a two-entity recognizer with {"FAKE_ACCOUNT": ["account", "bank"], "FAKE_ITIN": ["tax", "itin"]} boosts only the entity whose words appear nearby (pre-fix: no boost ever fires with a dict context — the 2 boost tests fail on main);
• configuration-path test: a YAML context dict reaches the constructed recognizer via RecognizerRegistryProvider.

Regression runs, all green on the branch:

• tests/test_per_entity_context_words.py — 8 passed
• tests/test_lemma_context_aware_enhancer.py, tests/test_pattern_recognizer.py, tests/test_recognizer_registry_provider.py, tests/test_context_support.py, tests/test_analyzer_engine.py — all passed
• uv run ruff check + ruff format — clean

Checklist

• I have reviewed the contribution guidelines
• I agree to follow this project's Code of Conduct
• I confirm that I have the right to submit this contribution and that it does not knowingly contain proprietary or confidential code.
• My code includes unit tests
• All unit tests and lint checks pass locally
• My PR contains documentation updates / additions if required