Apply shared term normalization to search terms, query matchers and the alias index #346

Follow-up to #322 / #325. #325 makes the memory and SQLite term and property indexes share normalize_term (NFC, lowercase, trim, collapse whitespace). Several other places still only call .lower(), so they can disagree with the index keys:

1. Search side. search.validate_and_prepare_term, searchlib._parse_search_term and query.get_matching_term_for_text (and the entity-matching helpers such as match_entity_name_or_type) still only lowercase. Example: for entity "Space Needle" and search term "space needle", lookup_term finds the semref, but boost_entities compares "space needle" with "space needle" and fails, so the entity loses its boost. Normalizing once in validate_and_prepare_term and in the query matchers would fix this.
2. Related-terms alias index (storage/memory/related_terms_index.py, storage/sqlite/related_terms_index.py) stores and looks up raw text. An alias registered for "space needle" is missed when the user searches " Space\tNeedle".
3. The two fuzzy-index update paths collect different terms. ConversationBase incremental updates add only entity names, topic text and action verbs. add_messages._collect_related_terms_for_fuzzy_index uses collect_entity_terms / collect_action_terms, which also include types, facets, subjects and objects. So related-term expansion depends on which ingestion path was used. This was already the case before Share term normalization between memory and SQLite indexes #325.

Found while reviewing #325; left out of it to keep that PR focused on the storage indexes.