feat: implement AI training ban and major CLI/Build pipeline overhaul

[LunaStev]

This PR introduces a strict "No-AI-Training" policy across the repository and executes a foundational redesign of the wavec CLI and compilation pipeline. By moving to a structured BuildPlan architecture and adding professional-grade compiler flags, Wave is now a much more capable toolchain for systems-level development and CI/CD integration.

---

Key Changes

1. AI/ML Training Prohibition & Policy Enforcement

• Explicit Policy: Added ai.txt to the repository root. This file explicitly prohibits the use of the codebase for crawling, scraping, pretraining, fine-tuning, or dataset construction without prior written permission.
• File Headers: Applied the AI TRAINING NOTICE header across all source files, including .rs, .wave, .py, .sh, and Dockerfile, to ensure the policy is visible at the source level.

2. Professional CLI & Build System Redesign

The CLI has been completely refactored from an imperative argument loop to a structured BuildRequest and BuildPlan architecture:

• Granular Output Control: Added the --emit=<kinds> flag. Supported outputs include ast, ir, bc, asm, obj, bin, and check.
• Fast Validation: Introduced the check command (alias for build --emit=check) to perform rapid syntax and semantic validation without the overhead of code generation.
• Advanced Linking & Compilation:
  • Added support for --shared, --static, --pie, --no-pie, and --freestanding.
  • Introduced low-level control with --entry, --linker-script, and --no-start-files.
• Multi-Format Input: The compiler can now process non-Wave files via --input-type=<wave|ir|bc|asm|obj>, allowing wavec to act as a wrapper for linking LLVM IR or assembly files.
• Introspection: Added the print command to query toolchain capabilities such as target-list, sysroot, cpu-list, and target-features.
• Dry-Run & JSON Formatting: Added a --dry-run mode and supported --error-format=json for better integration with IDEs and automation tools.

3. LLVM Backend & Runner Enhancements

• Model Configuration: Exposed code-model and relocation-model through the -C (Codegen) flags.
• Modular Frontend: Refactored src/runner.rs to expose modular functions like frontend_prepare_wave_ast and emit_wave_ast_text. This allows the compiler to handle complex multi-step build plans more efficiently.
• CI/CD Optimization: Updated CliError to return structured exit codes (1 for general, 2 for syntax/semantic, 3 for backend errors) to simplify pipeline logic.

---

Example Usage

Emitting LLVM IR and Assembly:

wavec build main.wave --emit=ir,asm

Rapid Semantic Checking:

wavec check main.wave

Linking with a custom linker script for a freestanding target:

wavec build kernel.wave --freestanding --linker-script link.ld --no-start-files

Querying supported targets:

wavec print target-list

---

Benefits

• Legal Protection: Explicitly protects the project's intellectual property from unauthorized AI ingestion.
• Toolchain Professionalism: The new --emit and linking flags bring wavec closer to the capabilities of clang or rustc.
• Pipeline Flexibility: Structured exit codes and JSON error formats make it significantly easier to use Wave in automated environments.