[RFC] Deprecate ti.cc backend in favor of TiRT & its C API?

[ailzhang]

What's ti.cc backend?

It's a backend that generates taichi kernels into C source files and then execute. It was the recommended way to deploy taichi kernels in an environment without Python.

Why are we thinking about deprecating it?

1. Taichi has converged to LLVM IR (for CPU/CUDA/AMD) and SPIR-V IR (for vulkan/metal/opengl/gles) as codegen targets and maintaining a separate C-source codegen has been pretty hard given the limited bandwidth we have. We weren't able to keep up the feature parity between ti.cc and other backends.
2. We believe TiRT along with its C API can fulfill the deployment needs where ti.cc was necessary, especially when it comes to using GPU resources (ti.cc runs purely on CPU). We've completed TiRT support for vulkan/opengl/metal backend and cpu/cuda support are experimental.

What's next?

Please let us know about your thoughts! Especially if you need ti.cc for a particular use case that TiRT cannot replace, please shout out loudly and we're happy to learn about it!
We're going to propose deprecating ti.cc backend in v1.5 release note to collect more feedback from users, and if everything goes well it'll be deprecated in v1.6 and removed around v1.7 (roughly mid 2023).

[ailzhang]

Closing as it's removed in #7836