Hyeongseok Oh

I don't know your big picture, but maybe you want to execute at once multiple executors. But I have concern `lowered graph` access and executors sharing `lowered graph` are thread-safe....

>>both const and non-const Please ignore this comment.

>IMHO, it can be done by checking input shapes of Bulk op in Static or Dynamic ShapeInferer. Pros of this approach is that there is no change on user's view....

Copy from `res/CircleSchema/0.4/circle_schema.fbs` and update to match TFLite 2.9.

Rebased

@nnfw-bot test - cross-runtime-test-release - x64-builder - cross-aarch64-runtime-test-release

I made this PR because tizen will update tensorflow version 2.9. But we can update this schema independently with tizen schedule. If we decide to update, we can update this...

Schema draft: #8442

Pending issue: x86-64 target is reverted to ubuntu 16.04. So we cannot use gcc-8 toolchain.

`luci-interpreter` is using TensorFlowRuySource 2.6 (`infra/cmake/packages/TensorFlowRuySource-2.6.0`), and runtime (ruy backend, cpu backend, and runtime core) will use TensorFlowRuySource 2.8 (`infra/cmake/packages/TensorFlowRuySource-2.8.0`). So it will be independent change with `luci-interpreter`.