feat: split global/shared memory words with S@/S!

[tetsuo-cpp]

Summary

• add S@ and S! words for explicit shared-memory load/store while keeping @/! for global memory
• add new dialect ops and parser mappings (forth.shared_load / forth.shared_store)
• lower shared memory ops through workgroup address-space pointers using gpu::AddressSpace::Workgroup
• update translation/conversion tests for the new words and shared address-space lowering
• update tiled matmul GPU test to use S@/S!

Test plan

• cmake --build build --target format
• uv run ruff check gpu_test/
• cmake --build build --target check-warpforth

Closes #36