fix: deduplicate entities by (title, type) to preserve same-name different-type entities

[octo-patch]

Fixes #1718

Problem

finalize_entities tracked only the entity title in its deduplication set, so two entities with the same title but different types (e.g. "Python" as LANGUAGE and "Python" as CONCEPT) were incorrectly collapsed into one. The second entity was silently dropped, causing ~10% data loss in graphs with polymorphic entity names.

Solution

Change the dedup key from title (a str) to (title, type) (a tuple[str, str]), matching the approach already used in finalize_relationships which deduplicates by (source, target).

Before:

seen_titles: set[str] = set()
...
if not title or title in seen_titles:
    continue
seen_titles.add(title)

After:

seen_entities: set[tuple[str, str]] = set()
...
entity_type = row.get("type", "")
if not title or (title, entity_type) in seen_entities:
    continue
seen_entities.add((title, entity_type))

Testing

• Updated the existing test_deduplicates_by_title docstring to clarify it covers same (title, type) duplicates.
• Added test_preserves_same_title_different_type which directly exercises the bug scenario: three entities where two share the title "Python" but have types LANGUAGE and CONCEPT; all three must survive finalization.
• All 32 existing tests continue to pass.