feat: complete Gate 1 source differential

[hiroki23]

Summary

• connect pinned TypeRB source projects to snapshot v2 through the removable, process-based reference producer
• add a TypeRB-authored application build driver and differential correctness suite
• complete checked Integer power and exact Go-baseline failure output/status
• inline common checked Integer operations after the first kernel measurement exposed a greater-than-2x regression
• add the TypeRB-authored reproducible benchmark harness, raw measurements, phase/RSS/toolchain accounting, and Gate 1 report
• run the full source differential corpus on pinned QBE 1.3 in macOS CI

Gate 1 result

Correctness passed for all supported projects, including exact stdout, stderr, and process status. Unsupported dynamic output fails explicitly.

Against the stronger applicable optimized Go baseline on the recorded Apple M2 Pro run:

• warm end-to-end build time improved by 30.5% to 36.3%
• stripped executable size improved by 96.85%
• the worst runtime result was a 16.0% regression, within the pre-registered 25% bound
• no final result crossed the 2x catastrophic limit

This passes Gate 1 and stops before Gate 2. It does not select QBE for production or represent final self-hosted compiler performance.

Verification

• pinned TypeRB formatting and checks for the compiler, benchmark tool, and every corpus project
• 18 TypeRB tests with QBE, including the complete source differential corpus
• 10 builds per candidate/project, 50 post-warmup runtime observations, raw and stripped size, phase time, peak RSS, dynamic dependencies, and toolchain size
• 726 recorded measurements, all with status 0

Closes #4