[REFACTOR][IR] Clean up PrimType follow-ups

[tqchen]

This PR follows up on the PrimType refactor by cleaning up a few residual type-boundary cases.

Summary:

• Use expression PrimType access for Metal TVMScript shuffle helpers.
• Extend namespace cleanup coverage for Metal shuffle up/down helpers.
• Simplify narrow backend Python raw dtype boundaries while leaving true raw ABI/codegen boundaries intact.
• Inline PrimType::StorageBytes() in the header and remove the exported out-of-line symbol.

Validation:

• git diff --check
• pre-commit run --files python/tvm/backend/metal/script.py tests/python/tirx/test_op_namespace_cleanup.py python/tvm/backend/cuda/operator/tile_primitive/elementwise/smem.py python/tvm/backend/trn/transform/naive_allocator.py
• pre-commit run --files include/tvm/ir/base_expr.h src/ir/type.cc
• cmake --build build --target tvm_runtime tvm_compiler -j$(nproc)
• python -m pytest tests/python/tirx/test_op_namespace_cleanup.py -xvs -rs (passes with 10 skips due the suite's sm_100a device gate)
• checked libtvm_compiler.so no longer exports tvm::PrimType::StorageBytes

[gemini-code-assist]

Code Review

This pull request inlines the StorageBytes() method of PrimType in include/tvm/ir/base_expr.h, removes its out-of-line definition in src/ir/type.cc, and cleans up unused DataType imports and dtype property accesses in Python backends. It also updates Metal SIMD shuffle intrinsics to use var.ty instead of var.dtype and adds corresponding unit tests. Regarding the feedback, the implementation of StorageBytes() contains a redundant check for 1-bit, 1-lane kDLUInt that can be removed since the general formula already handles this case correctly.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

The special case check for kDLUInt with 1 bit and 1 lane is redundant. The general formula (bits * lanes + 7) / 8 already correctly evaluates to 1 when bits == 1 and lanes == 1 (i.e., (1 * 1 + 7) / 8 = 1). Removing this check simplifies the code and avoids unnecessary branching.

    return static_cast<size_t>(
        (static_cast<uint64_t>(dtype.bits) * static_cast<uint64_t>(dtype.lanes) + 7) / 8);