docs: restructure CodebaseOverview "Query Entry Points" to lead with ask_knowledge_base

Context (A2A — Antigravity, Claude Opus 4.6)

#9893 successfully promoted ask_knowledge_base across AGENTS.md (§2.1 tool hierarchy table) and AGENTS_STARTUP.md (§3.4 "Ask the Expert" protocol). However, the CodebaseOverview — the first document every agent reads on startup — still structurally privileges query_documents over ask_knowledge_base.

The Structural Imbalance

The "Key Concepts: Query Entry Points" section (L538-590) currently allocates:

• 1 bullet (~1 line) to ask_knowledge_base (L542)
• 45+ lines to query_documents keyword lists organized into 10 subsections (L543-590)

Agents scanning this section will naturally gravitate to the visually dominant tool. The hierarchy is inverted.

Empirical Proof (Same Query, Two Tools)

Tested with: "How does the Core Engine pipeline work?"

The ask_knowledge_base tool is functionally a zero-cost RAG subagent — it reads, retrieves, and synthesizes answers from the current indexed codebase. For conceptual questions, it eliminates the query_documents → view_file → read → interpret chain entirely.

Proposed Change

Restructure the "Key Concepts: Query Entry Points" section to:

1. Lead with ask_knowledge_base — expand from 1 line to 5-8 lines with 3-4 example queries showing the tool's power for different question types (conceptual, syntax verification, architectural patterns)
2. Add a clear decision table — when to use ask vs query_documents vs view_file (similar to the hierarchy table in AGENTS.md §2.1, but adapted for the CodebaseOverview's pedagogical style)
3. Reposition query_documents keyword lists — keep them as a secondary reference under a subheading like "File Discovery Keywords" rather than as the section's primary content
4. Add anti-pattern warning — "Do NOT attempt to answer Neo.mjs questions from training data. The framework evolves rapidly — your training data is almost certainly outdated for syntax-level details."

Content Sketch

<h2 class="neo-h2" data-record-id="7">Key Concepts: Query Entry Points</h2>

<h3 class="neo-h3" data-record-id="8">Ask First, Search Second</h3>

Your most powerful tool is `ask_knowledge_base`. It reads multiple source files, 
synthesizes a grounded answer, and cites its references — all in a single call.

**Example queries:**
- `ask_knowledge_base(query='how does the reactive config system work?')`
- `ask_knowledge_base(query='current syntax for state provider bindings')`
- `ask_knowledge_base(query='how to define a Grid with multiple bodies')`
- `ask_knowledge_base(query='what is the difference between ntype and className?', type='guide')`

| Question Type | Tool | Returns |
|---|---|---|
| "How does X work?" | `ask_knowledge_base` | Synthesized answer + source citations |
| "Which files implement Z?" | `query_documents` | Ranked file paths with relevance scores |
| Exact implementation details | `view_file` | Raw source code |

<h3 class="neo-h3" data-record-id="9">File Discovery Keywords</h3>

When you specifically need to discover *which files* implement something, 
use `query_documents` with these high-value terms:
[...existing keyword lists...]

Scope

• Documentation-only. No code changes.
• Single file modified: learn/guides/fundamentals/CodebaseOverview.md
• Builds on #9893 (already closed — promoted ask_knowledge_base in AGENTS.md and AGENTS_STARTUP.md)
• Builds on #9898 (already closed — added Core Engine cross-references and tree.json pointers)

Architectural Context

• The CodebaseOverview is read in AGENTS_STARTUP.md Step 1 — it's the first document agents see
• ask_knowledge_base is backed by the Knowledge Base MCP server's LearningSource.mjs which uses tree.json to index all 130+ guides
• The tool can filter by content type (guide, src, example, ticket, blog, release, test)