[OpenCL] Refactor OpenCL runtime to support SPIRV binary ingestion

[masahi]

This is the first step toward supporting Intel XMX (tensorcore equivalent of Intel dGPUs). They expose XMX via spirv, and such spirv modules can be consumed by their OpenCL runtime. It is going to be a long game, since their extension is not even in the spirv toolchain yet. I have the simplest test case of XMX tensorization (m8n8k16 matmul) working using this script but beyond that I've been struggling with strange correctness issues (see intel/intel-graphics-compiler#279).

For now, I'm abusing the existing -device target option to decide when to enable spirv code gen for the opencl target (target = "opencl -device=spirv"). I don't have a strong opinion on this, suggestions welcome.

The next steps are going to be

• Update our spirv codegen to support emitting the OpenCL flavor ("Kernel" capability, https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_Env.html)
• Support Intel joint matrix extension in our spirv codegen. It's very similar to the NV tensorcore extension [Vulkan] Add VK_NV_cooperative_matrix support #14770

@echuraev @tqchen @csullivan

[tvm-bot]

Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.

• cc @echuraev, @elvin-n _{See #10317 for details}

_{Generated by tvm-bot}

[echuraev]

In general LGTM. Several minor comments.

[echuraev]

LGTM. Thank you!