Add input sanitization for prompts and context sent to LLMs

[sfloess]

Part of #32: Security Hardening - API Key Storage and Input Sanitization

Scope

Sanitize user prompts and code context before sending to LLM APIs to prevent prompt injection and information leakage.

What to do

• Identify all places where user input or code context is assembled into API requests
• Strip or escape potentially dangerous patterns (e.g., system prompt overrides, delimiter injection)
• Ensure file paths, environment variables, and other system information are not inadvertently included in context
• Add configurable context size limits to prevent sending entire large files
• Consider a "review before send" option for sensitive contexts

Acceptance criteria

• User input is sanitized before inclusion in API requests
• No system environment information leaks into prompts unintentionally
• Context size is bounded to prevent excessive data transmission
• Sanitization is centralized (not duplicated across 9 modules)