Add operations for GCC builtin functions

[Lancern]

Similar to the LLVMIR dialect, we should provide dedicated operations for each GCC builtin function.

[SchrodingerZhu]

Each is too much.

There are thousands of them and many of them are target specific. I think we should only consider most useful categories such as alloca/vector/overflow+arith/atomic.

For vector, we may also need to emit some always_inline functions to facilitate the translation, such as unaligned loading or reduce. Fortunately, rustc codegen project provides some examples there.

For aligned alloca and other stub builtins, we need to wait for upstream to update their support. See antoyo/libgccjit#5. I think we can release the first beta which can help us to push forward related patches in their IRC/ML.

[Lancern]

Each is too much.

OK, then we might cover those frequently used builtins first. LLVMIR dialect does not cover all LLVM intrinsics either.

Anyway let's remove this issue from the initial release milestone as this could be a long-term development process.