docs: Fix documentation accuracy and add missing Gemini provider

Corrections made:
- docs/architecture.md: Fix crate count from 9 to 8 (actual count)
- docs/tools.md: Fix code_search supported languages (kotlin -> haskell, scheme, racket)
- docs/CODE_SEARCH.md: Add missing Haskell and Scheme to supported languages list
- docs/providers.md: Add complete Gemini provider documentation section
- docs/configuration.md: Add Gemini configuration section

The Gemini provider (crates/g3-providers/src/gemini.rs) was fully implemented
but not documented. The code_search tool actually supports haskell and scheme
(via tree-sitter) but documentation incorrectly listed kotlin.

Agent: lamport

docs/CODE_SEARCH.md

@@ -40,8 +40,9 @@ g3 includes a syntax-aware code search tool powered by tree-sitter. Unlike text-
 - Java
 - C
 - C++
-- Racket
+- Haskell
 - Scheme
+- Racket
 
 ## Basic Usage
 

docs/architecture.md

@@ -57,7 +57,7 @@ g3 follows a **tool-first philosophy**: instead of just providing advice, it act
 
 ## Workspace Structure
 
-g3 is organized as a Rust workspace with 9 crates:
+g3 is organized as a Rust workspace with 8 crates:
 
 ```
 g3/

docs/configuration.md

@@ -89,6 +89,21 @@ use_oauth = true                 # Use OAuth (recommended)
 - **OAuth** (`use_oauth = true`): Opens browser for authentication, tokens refresh automatically
 - **Token** (`token = "..."`, `use_oauth = false`): Uses personal access token directly
 
+### Gemini Configuration
+
+```toml
+[providers.gemini.default]
+api_key = "your-google-api-key"  # Required: Your Google AI API key
+model = "gemini-2.0-flash"       # Model to use
+max_tokens = 8192
+temperature = 0.7
+```
+
+**Available Gemini models**:
+- `gemini-2.0-flash` (recommended)
+- `gemini-1.5-pro`
+- `gemini-1.5-flash`
+
 ### OpenAI Configuration
 
 ```toml

docs/providers.md

@@ -1,6 +1,6 @@
 # g3 LLM Providers Guide
 
-**Last updated**: January 2025  
+**Last updated**: January 2025 (Gemini provider added)  
 **Source of truth**: `crates/g3-providers/src/`
 
 ## Purpose
@@ -13,6 +13,7 @@ This document describes the LLM providers supported by g3, their capabilities, a
 |----------|------|--------------|---------------|----------------|----------|
 | **Anthropic** | Cloud | Native | Yes | 200k (1M optional) | General use, complex tasks |
 | **Databricks** | Cloud | Native | Yes (Claude models) | Varies | Enterprise, existing Databricks users |
+| **Gemini** | Cloud | Native | No | 1M-2M | Google ecosystem, large context |
 | **OpenAI** | Cloud | Native | No | 128k | GPT model preference |
 | **OpenAI-Compatible** | Cloud | Native | No | Varies | OpenRouter, Groq, Together, etc. |
 | **Embedded** | Local | JSON fallback | No | 4k-32k | Privacy, offline, cost savings |
@@ -121,6 +122,50 @@ Models depend on your Databricks workspace configuration:
 
 ---
 
+## Gemini
+
+**Location**: `crates/g3-providers/src/gemini.rs`
+
+### Features
+
+- **Native tool calling**: Full support for structured tool calls
+- **Large context windows**: Up to 2M tokens on some models
+- **Streaming**: Real-time response streaming
+- **Google ecosystem**: Integrates with Google Cloud
+
+### Configuration
+
+```toml
+[providers.gemini.default]
+api_key = "your-google-api-key"  # Required
+model = "gemini-2.0-flash"       # Model name
+max_tokens = 8192                # Max output tokens
+temperature = 0.7                # 0.0-1.0
+```
+
+### Available Models
+
+| Model | Context | Notes |
+|-------|---------|-------|
+| `gemini-2.0-flash` | 1M | Fast, efficient |
+| `gemini-1.5-pro` | 2M | Most capable |
+| `gemini-1.5-flash` | 1M | Balanced speed/quality |
+
+### Getting an API Key
+
+1. Go to [Google AI Studio](https://aistudio.google.com/)
+2. Create or select a project
+3. Generate an API key
+4. Add to your g3 configuration
+
+### Notes
+
+- Gemini models have very large context windows (1M-2M tokens)
+- Good for tasks requiring extensive context
+- Native tool calling works well for agentic workflows
+
+---
+
 ## OpenAI
 
 **Location**: `crates/g3-providers/src/openai.rs`
@@ -291,6 +336,7 @@ This works but is less reliable than native tool calling.
 | Privacy-critical | Embedded |
 | Offline development | Embedded |
 | Fast iteration | Groq (Llama) |
+| Large context needs | Gemini (1M-2M tokens) |
 | Model variety | OpenRouter |
 
 ### By Priority

docs/tools.md

@@ -233,7 +233,7 @@ Syntax-aware code search using tree-sitter.
 - `max_concurrency` (integer, optional): Parallel searches (default: 4)
 - `max_matches_per_search` (integer, optional): Max matches (default: 500)
 
-**Supported languages**: rust, python, javascript, typescript, go, java, c, cpp, kotlin
+**Supported languages**: rust, python, javascript, typescript, go, java, c, cpp, haskell, scheme, racket
 
 **Example**:
 ```json