[FFI] Update the interface of ffi.load_inline to match torch

[yaoyaoding]

This PR update the interface of ffi.load_inline to match torch.utils.cpp_extensions.load_inline:

1. Rename cpp_source to cpp_sources, cuda_source to cuda_sources.
2. Unify the cpp_functions and cuda_functions into functions.
3. Add build_directory to allow the user to specify the build directory directly.

Example:

    mod: Module = tvm_ffi.cpp.load_inline(
        name="hello",
        cpp_sources=r"""
            void add_one_cpu(DLTensor* x, DLTensor* y) {
              // implementation of a library function
              TVM_FFI_ICHECK(x->ndim == 1) << "x must be a 1D tensor";
              DLDataType f32_dtype{kDLFloat, 32, 1};
              TVM_FFI_ICHECK(x->dtype == f32_dtype) << "x must be a float tensor";
              TVM_FFI_ICHECK(y->ndim == 1) << "y must be a 1D tensor";
              TVM_FFI_ICHECK(y->dtype == f32_dtype) << "y must be a float tensor";
              TVM_FFI_ICHECK(x->shape[0] == y->shape[0]) << "x and y must have the same shape";
              for (int i = 0; i < x->shape[0]; ++i) {
                static_cast<float*>(y->data)[i] = static_cast<float*>(x->data)[i] + 1;
              }
            }
        """,
        functions=["add_one_cpu"],
    )

    x = numpy.array([1, 2, 3, 4, 5], dtype=numpy.float32)
    y = numpy.empty_like(x)
    mod.add_one_cpu(x, y)
    numpy.testing.assert_equal(x + 1, y)

[tqchen]

lgtm modulo lint