docs: remove deprecated Graph Memory references

Graph Memory was deprecated on April 16, 2026. This removes remaining
mentions across docs that could mislead readers into thinking it is still
a supported feature.

- Platform pages (overview, quickstart, faqs, platform-vs-oss, vibecoding):
  remove graph memory from feature lists, comparison table, and intro copy
- Platform features (platform-overview, mcp-integration,
  advanced-memory-operations): rephrase "graph-powered retrieval" to
  "advanced retrieval"; drop "Enable graph memories" tip; remove graph-edge
  verification step from the metadata example
- Core concepts (add, search, memory-types, memory-evaluation): remove
  "optional graph storage" / "graph features" / "graph writes" / "graph
  toggles"; drop "graph" from Entity Search description; remove outdated
  add/search architecture diagrams (replacement pending)
- Open source features (overview, rest-api): remove "graph relationships"
  and "graph backend(s)" from REST server config narrative
- API reference (organizations-projects): drop "graph settings" from
  Update Project Settings description
- Cookbooks: remove "Graph Memory on Neptune" card (page removed) and
  Neptune mention from AWS Bedrock card; remove "Graph Capabilities"
  bullet from Gemini cookbook; remove graph-memory conditional bullet from
  controlling-memory-ingestion
- CLI: remove --graph / --no-graph flags from mem0 add / mem0 search
  flag tables and MEM0_ENABLE_GRAPH env var
- Images: delete orphan add_architecture.png and search_architecture.png

docs/core-concepts/memory-evaluation.mdx

@@ -47,7 +47,7 @@ When a query arrives, the retrieval pipeline scores candidates across three sign
 
 1. **Semantic Search** — Vector similarity scoring against memory embeddings
 2. **Keyword Search** — Normalized term matching via BM25 with verb-form lemmatization
-3. **Entity Search** — Entity graph matching boosts memories linked to query entities
+3. **Entity Search** — Entity matching boosts memories linked to query entities
 
 Results are fused via rank scoring into a final top-K set. Different query types lean on different signals: