refactor(g3-cli): break lib.rs into focused modules

Extract 7 modules from the 2966-line lib.rs:
- cli_args.rs (133 lines): CLI argument parsing with clap
- autonomous.rs (785 lines): coach-player feedback loop
- agent_mode.rs (284 lines): specialized agent execution
- accumulative.rs (343 lines): iterative requirements mode
- interactive.rs (851 lines): REPL with command handling
- task_execution.rs (212 lines): unified retry logic
- utils.rs (91 lines): display and workspace helpers

Key improvements:
- lib.rs reduced from 2966 to 415 lines (86% reduction)
- Eliminated duplicate retry logic between execute_task and execute_task_machine
- Each module has a single responsibility
- Easier to reason about and maintain

Agent: fowler

crates/g3-cli/src/cli_args.rs

@@ -0,0 +1,133 @@
+//! CLI argument parsing for G3.
+
+use clap::Parser;
+use std::path::PathBuf;
+
+#[derive(Parser, Clone)]
+#[command(name = "g3")]
+#[command(about = "A modular, composable AI coding agent")]
+#[command(version)]
+pub struct Cli {
+    /// Enable verbose logging
+    #[arg(short, long)]
+    pub verbose: bool,
+
+    /// Enable manual control of context compaction (disables auto-compact at 90%)
+    #[arg(long = "manual-compact")]
+    pub manual_compact: bool,
+
+    /// Show the system prompt being sent to the LLM
+    #[arg(long)]
+    pub show_prompt: bool,
+
+    /// Show the generated code before execution
+    #[arg(long)]
+    pub show_code: bool,
+
+    /// Configuration file path
+    #[arg(short, long)]
+    pub config: Option<String>,
+
+    /// Workspace directory (defaults to current directory)
+    #[arg(short, long)]
+    pub workspace: Option<PathBuf>,
+
+    /// Task to execute (if provided, runs in single-shot mode instead of interactive)
+    pub task: Option<String>,
+
+    /// Enable autonomous mode with coach-player feedback loop
+    #[arg(long)]
+    pub autonomous: bool,
+
+    /// Maximum number of turns in autonomous mode (default: 5)
+    #[arg(long, default_value = "5")]
+    pub max_turns: usize,
+
+    /// Override requirements text for autonomous mode (instead of reading from requirements.md)
+    #[arg(long, value_name = "TEXT")]
+    pub requirements: Option<String>,
+
+    /// Enable accumulative autonomous mode (default is chat mode)
+    #[arg(long)]
+    pub auto: bool,
+
+    /// Enable interactive chat mode (no autonomous runs)
+    #[arg(long)]
+    pub chat: bool,
+
+    /// Enable machine-friendly output mode with JSON markers and stats
+    #[arg(long)]
+    pub machine: bool,
+
+    /// Override the configured provider (anthropic, databricks, embedded, openai)
+    #[arg(long, value_name = "PROVIDER")]
+    pub provider: Option<String>,
+
+    /// Override the model for the selected provider
+    #[arg(long, value_name = "MODEL")]
+    pub model: Option<String>,
+
+    /// Disable log file creation (no logs/ directory or session logs)
+    #[arg(long)]
+    pub quiet: bool,
+
+    /// Enable WebDriver browser automation tools
+    #[arg(long)]
+    pub webdriver: bool,
+
+    /// Use Chrome in headless mode for WebDriver (instead of Safari)
+    #[arg(long)]
+    pub chrome_headless: bool,
+
+    /// Use Safari for WebDriver (this is the default)
+    #[arg(long)]
+    pub safari: bool,
+
+    /// Enable flock mode - parallel multi-agent development
+    #[arg(long, requires = "flock_workspace", requires = "segments")]
+    pub project: Option<PathBuf>,
+
+    /// Flock workspace directory (where segment copies will be created)
+    #[arg(long, requires = "project")]
+    pub flock_workspace: Option<PathBuf>,
+
+    /// Number of segments to partition work into (for flock mode)
+    #[arg(long, requires = "project")]
+    pub segments: Option<usize>,
+
+    /// Maximum turns per segment in flock mode (default: 5)
+    #[arg(long, default_value = "5")]
+    pub flock_max_turns: usize,
+
+    /// Enable planning mode for requirements-driven development
+    #[arg(long, conflicts_with_all = ["autonomous", "auto", "chat"])]
+    pub planning: bool,
+
+    /// Path to the codebase to work on (for planning mode)
+    #[arg(long, value_name = "PATH")]
+    pub codepath: Option<String>,
+
+    /// Disable git operations in planning mode
+    #[arg(long)]
+    pub no_git: bool,
+
+    /// Enable fast codebase discovery before first LLM turn
+    #[arg(long, value_name = "PATH")]
+    pub codebase_fast_start: Option<PathBuf>,
+
+    /// Run as a specialized agent (loads prompt from agents/<name>.md)
+    #[arg(long, value_name = "NAME", conflicts_with_all = ["autonomous", "auto", "chat", "planning"])]
+    pub agent: Option<String>,
+
+    /// Skip session resumption and force a new session (for agent mode)
+    #[arg(long)]
+    pub new_session: bool,
+
+    /// Automatically remind LLM to call remember tool after turns with tool calls
+    #[arg(long)]
+    pub auto_memory: bool,
+
+    /// Enable aggressive context dehydration (save context to disk on compaction)
+    #[arg(long)]
+    pub acd: bool,
+}