v0.8.0

This release is a monster. If upgrading from 0.7.x, I recommend starting with a fresh config file. There are several breaking changes to the config. See updated docs linked in the readme.

**Breaking Changes:**
- The `web_port` is no longer configurable and is fixed at 5000
- For RTMP support, port 1935 needs to be made available
- All previous automations built around `frigate/<camera>/events/start|end` will need to be replaced
- The clips_dir is now at `/media/frigate/clips` and the cache_dir now at `/tmp/cache`. These are no longer configurable.
- `/debug/stats` is now `/api/stats`
- If you are specifying your thresholds for person detection, 0.85 will now be WAY too high. Update them to 0.7 or remove from your config and use the default. The new model doesn’t score detections as high, but has a much lower false positive rate.
- The previous `mask` value is now under the `motion` config section
- Masks are now configured like zones and image masks are no longer supported
- Fallback to CPU if the EdgeTPU is not detected has been disabled. It will now keep trying to detect your EdgeTPU.

**Changes:**
- New [HomeAssistant integration](https://github.com/blakeblackshear/frigate-hass-integration) should replace all of your previous MQTT and REST sensors
- Up to 4 unique streams can be provided for each camera and assigned the following roles:
  - detect: stream used for object detection
  - clips: stream used for clips
  - record: stream used for 24/7 recording
  - rtmp: stream re-broadcasted by frigate as a RTMP feed for homeassistant
- 24/7 recording of video streams with retention management
- Clips are maintained in a local database and served up via an api
- Clip retention policy (per camera/object)
- Complete revamp of MQTT event messages for easy notification setup
- Improved calculation for best images
- MQTT topics now publish object type counts rather than just on/off
- Support for mesa va drivers for AMD hwaccel in ffmpeg
- New image with nvidia acceleration for ffmpeg decoding (this will not use your GPU for object detection)
- Config validation at startup
- Ability to configure model dimensions
- Handle portrait cameras better
- Update model to mobiledet and update default threshold to .7
- Make motion detection configurable
- Make object tracking configurable
- Add a version endpoint
- Mask can now be a list of masks
- Mjpeg and latest.jpg now support dynamic drawing of bounding boxes, regions, zones, motion, mask, and timestamp via query string parameters
- Update crop of thumbnail based on object box rather than detection region
- `post_capture` setting for clips
- Allow `num_threads` for CPU detectors
- Add `qsv` hwaccel for ffmpeg (should help 10th gen intel processors)
- Updates for events on zone changes
- A basic web UI
- Environment variables can be set via the config file
- Optionally save snapshot jpg files alongside clips
- New `mqtt` options for configuring the images sent via MQTT
- Disable/enable detection, clips, snapshots via MQTT (retained in MQTT to persist setting on restart)
- Stats also now published via MQTT
- Object detection masks by object type
- Version and uptime added to stats

Images:
- x86: `blakeblackshear/frigate:0.8.0-amd64`
- x86 w/ nvidia (for ffmpeg only): `blakeblackshear/frigate:0.8.0-amd64nvidia`
- Raspberry Pi 3/4 (32bit): `blakeblackshear/frigate:0.8.0-armv7`
- Raspberry Pi 4 (64bit): `blakeblackshear/frigate:0.8.0-aarch64`