[WLDFA]

Summary

• add TileArrayOutputRef binary producer overloads for (SubTileView, ordinary Tile) and (ordinary Tile, SubTileView)
• emit the existing source-side B.SUBVIEW 0/1 binding together with the destination B.ASSEMBLE INIT/MIDDLE/LAST lifecycle
• preserve CUBE_M32 layout and validate dtype, location, physical/valid shape, fragment capacity, writer size, and subview size
• cover both source orders across all 12 binary elementwise wrappers

Tests

• compiled the issue [TileOP API] Support mixed SubTileView and ordinary Tile sources for assembled binary ops #299 reproduction with the target Linx Clang to syntax, LLVM IR, and object code
• verified one source B.SUBVIEW per producer with selector 0/1 matching the source order
• verified eight-slot INIT + six MIDDLE + LAST assembly, source offsets, destination offsets, parent SizeCode 4, and fragment SizeCode 1
• verified all 12 binary operations instantiate in both mixed orders
• verified expected diagnostics for layout, dtype, physical shape, valid shape, location, and capacity mismatches
• git diff --check
• bash -n test/tileop_api/run_negatives.sh test/tileop_api/verify_target_cxx_frontend.sh
• full verify_target_cxx_frontend.sh reached the pre-existing unrelated failure FAIL: Shared handle materialized to memory after the new issue [TileOP API] Support mixed SubTileView and ordinary Tile sources for assembled binary ops #299 checks passed

Fixes #299