[BUG] merge_entities=True merges entities with different id, type AND name (Person 'Alice' + Organization 'Acme' collapse into one) #1137

Bug Description

With merge_entities=True and default settings, GraphBuilder merges two entities that share
no identifier, no type and no name. A Person named "Alice" and an Organization
named "Acme" are collapsed into a single entity, and the Organization is lost.

merge_entities=True is the value used in the Neo4j example in docs/quickstart.md, so this is
reachable by following the documentation.

This is distinct from #1085 / #1110 / #1111, which concern preserving entity_id aliases and
remapping endpoints when merging entities that genuinely are duplicates. Here the problem is
upstream of that: two entities that are not duplicates by any field are grouped as duplicates.

Steps to Reproduce

No graph store required — reproduces purely in memory:

from semantica.kg import GraphBuilder

entities = [
    {"id": "e1", "type": "Person",       "text": "Alice", "name": "Alice"},
    {"id": "e2", "type": "Organization", "text": "Acme",  "name": "Acme"},
]
relationships = [{"source_id": "e1", "target_id": "e2", "type": "knows"}]

for merge in (False, True):
    g = GraphBuilder(merge_entities=merge).build(
        {"entities": entities, "relationships": relationships}
    )
    ents = g.get("entities") or g.get("nodes") or []
    print(merge, len(ents), [e.get("name") or e.get("text") for e in ents])

Expected Behavior

False 2 ['Alice', 'Acme']
True  2 ['Alice', 'Acme']

Two entities differing in id, type and name are not duplicates under any reasonable
similarity threshold.

Actual Behavior

False 2 ['Alice', 'Acme']
True  1 ['Alice']

The Organization entity is silently dropped. Progress output shows
DuplicateDetector creating one candidate pair and MergeStrategyManager merging via
keep_most_complete, so the pair is being scored as a duplicate rather than discarded by a
threshold.

Persisting the same input to Neo4j writes 1 node instead of 2, which is how this was first noticed.

Environment

• OS: macOS (Darwin 25.6.0, arm64)
• Python Version: 3.12.14
• Semantica Version: 0.6.5 (origin/main @ 2a303cf)
• Installation Method: pipx (pipx install "semantica[graph-neo4j]")

Notes / possible directions

• A type guard would catch this specific case cheaply: entities with different type values are
  rarely duplicates, and refusing to merge across types would be a conservative default.
• Worth checking what similarity the default DuplicateDetector actually assigns to this pair —
  if it is above threshold, the scoring may be dominated by a field that is empty in both
  (both entities here have no properties), which would make short/sparse entities merge readily.
• Silently dropping an entity is a strong default. Even with merging enabled, reporting what was
  merged into what (or requiring an explicit strategy) would make the behaviour auditable.

Happy to test a patch against this reproduction.