Split backend crate into logical components

[avrabe]

Goal

Split synth-backend (3,800 lines) into focused crates.

Proposed Structure

1. synth-codegen: ARM instruction encoding

   • arm_encoder.rs
   • Instruction-level operations

2. synth-binary: Binary emission

   • elf_builder.rs
   • linker_script.rs
   • Binary file generation

3. synth-target: Target-specific features

   • vector_table.rs
   • reset_handler.rs
   • mpu.rs, arm_startup.rs
   • Memory Protection Unit

Tasks

• Create three new crates
• Split backend code appropriately
• Update inter-crate dependencies
• Migrate tests
• Update documentation

Acceptance Criteria

• ✅ Three separate crates with clear boundaries
• ✅ All tests pass
• ✅ Cleaner compilation pipeline

Dependencies

• Requires: Issue Consolidate root-level documentation #5 (audit)

[avrabe]

Closing based on crate audit (#9). The analysis shows synth-backend is already well-modularized internally. Splitting would add unnecessary complexity. No split needed.