Tensorneko

Features:
- Add extension argument for image browser server

Fix:
- The bug that cannot use image browser server
- Constrain the `lightning` version as `1.7.*`

Features:
- `FID` now can be runned in other devices than cpu
- `circular_pad` for list
- `RoundRobinDataset` for sampling data from multiple datasets
- `FaceXZooFaceDetector` now can be downloaded automatically via `install` method.
- `image_browser` server to build a web based remote image browser

Fix:
- Fix fail import in Python 3.7
- Remove requirements that pytorch < 1.12
- Precision issue of `Timer`

Features:
- `no_blocking` context manager for eventbus
- Better debugging traceback for `Conv`, `Linear` and patching layers.
- Remove `tqdm` and `matplotlib` from dependencies
- Add `frame2video` ffmpeg command
- Move `crop_with_padding` applied on `ndarray` to `tensorneko_util`\
- `FaceXZooFaceDetector` as a face detector based on [`FaceXZoo`](https://github.com/JDAI-CV/FaceX-Zoo/tree/main/face_sdk) repository. Here the [forked version](https://github.com/ControlNet/FaceX-Zoo) is used.

Fix:
- bug when using `frames2video` with torchvision backend.

Features:
- `sparse2binary` and `binary2sparse` to process labels to convert between categorical sparse vectors with dense binary vectors.
- Multi-backend `rgb2gray` to convert rgb images as grayscale
- Out-of-box face detector `OpencvFaceDetector`

Features:
- `dispatch.base` decorator used to implement multidispatch function in different py modules.
- `Eval` monad for evaluating variable by-name/by-need/by-value.

Fix:
- The address port is occupied when starting `Server` in Linux.

Features:
- `clip_image` with frame indexes or range for video via FFmpeg
- add h5 support for `io`
- impl `Seq.length` and `Seq.size` to get length of the sequence
- Now `backend` module is imported automatically