Add CLI and config file support for LLM timeout settings

[Iron-Ham]

Summary

Adds user-configurable timeout and retry settings for LLM requests. Users can now tune these values to better match their system's performance characteristics.

New options:

Setting	CLI Flag	Config Key	Default
Timeout	--timeout	llmTimeout	30s
Max attempts	--max-attempts	llmMaxAttempts	3

Precedence: CLI flags override config file settings, which override defaults.

Example usage:

# Single invocation with custom timeout
scg generate --timeout 60 --max-attempts 5

# Persist in config file (~/.config/scg/config.json on Linux, ~/Library/Application Support/scg/config.json on macOS)
{
  "llmTimeout": 60,
  "llmMaxAttempts": 5
}

Changes

• UserConfiguration.swift: Add llmTimeout: Int? and llmMaxAttempts: Int? fields
• CommitGenOptions.swift: Add llmTimeout: TimeInterval? and llmMaxAttempts: Int? fields
• GenerateCommand.swift: Add --timeout and --max-attempts CLI options with resolution logic
• CommitGenTool.swift: Create FoundationModelsClient with custom configuration from options

Stacking

This PR is stacked on #20 (LLM timeout enforcement). Merge #20 first.

Test plan

• Build passes with swift build
• All 49 existing tests pass
• scg generate --help shows new options
• Manual test: --timeout 10 uses shorter timeout
• Manual test: config file llmTimeout is respected

Fixes #18