Tensorneko

Features:
- `iou_2d` to evaluate IoU score for 2D areas.

Fix:
- Auto log metrics in `NekoModel.test_step`
- Fix the wrong implementation of positional encoding and class token in `TransformerEncoderBlock`

Features:
- Add `Seq.sort` method, to sort the sequence by itself or any given key.
- Add `progress_bar` and `parallel_type` argument for `Stream.for_each` and `Stream.with_for_each` to support progress bar and run in parallel.

Features:
- Add `parallel_type` argument for `Seq.map` and `Seq.for_each`. The default value is `None`. When one `ParallelType` is assigned to this argument, the program will run this step with multiple processes or threads.
- Add `Seq.with_for_each` return self but with running `for_each`.

Features:
- `for_each` methods for `AbstractSeq` (`Seq` and `Stream`).
- Add `progress_bar` argument for `Seq.map` and `Seq.for_each`. It will use `tqdm` progress bar.

Features:
- Add default implementation for `predict_step` and `test_step` in `NekoModel`
- Add `parallel_map` for `Seq` with 2 types of parallel methods, thread and process.

Fix:
- Add constrain to `torchmetrics` as the compatibility issue of `pytorch_lightning`.

Features:
- Refactor the `fn.py` library to integrate into the project because `fn.py` has a long time not been maintained, which is not compatible with Python 3.10.
- `tensorneko.util.Option`, `Some` and `Empty`: A monad to handle optional data.
- `tensorneko.util.Seq`: A linear collection able to process in the pipeline with method chaining.
- `tensorneko_util` library now supports Python 3.10.