Implement environment fingerprinting BLE positioning

[mvgijssel]

Mentioned here https://github.com/mvgijssel/setup/issues/139#issuecomment-1383199097 the accuracy of the trilateration algorithm isn't really high so going to investigate alternatives. One of those alternatives is to use something like environment fingerprinting with a machine learning regression (for example https://www.mdpi.com/2079-9292/9/1/44).

To get started with this try out https://github.com/schollz/find3 which already implements training, testing and different machine learning classifiers.

TODO

• [ ] train find3 in living_room
• [ ] train find3 in kitchen
• [ ] train find3 in office
• [ ] train find3 in hallway + toilet
• [ ] train find3 in landing + laundry room
• [ ] train find3 in baby_room
• [ ] train find3 in bedroom
• [ ] tain find3 in bathroom
• [x] Remove ESPresence IPS AppDaemon and nimble_distance component from shelly and moko devices
• [x] Setup mqtt sensor in home assistant which takes the probability into considered when changing locations. Only change on > 50% probability?
• [x] Only allow the location mqtt sensor to change after a certain time? To reduce jitter? Let's try to require 2x the reporting time of find3, so 2x2 = 4 seconds of reporting a value is required before turning to that value.
• [x] Test is away mode works by going outside
• [x] Change sensor icon when away

[mvgijssel]

The toilet is inside the hallway, but because it's a relative small area there's a lot of inaccuracy whether an object is within the hallway or toilet. To compensate we can change the logic as follows:

1. Consider the toilet to be a part of the hallway, it's no longer a separate location
2. The toilet will only toggle occupancy due to the motion sensor (not due to the object tracking). Can be achieved with https://github.com/mvgijssel/setup/issues/110
3. Once the toilet is occupied it will remain occupied until the occupancy of the hallway is disabled

Consider the same logic for the landing + laundry_room, which is similar to the toilet relatively small, resulting in accuracy.

[mvgijssel]

Need to re-train the locations due to the ESPHome devices becoming unreachable after a while impacting the fingerprinting. Fix is in https://github.com/mvgijssel/setup/pull/162.

[mvgijssel]

Updated the training settings to be as follows:

{
    "family":"home",
    "location": "office",
    "device":"bluetooth-00:00:00:00:00:00",
    "minimum_passive": -1,
    "window": 3
}

The window is now 3 seconds for collecting (2.5 times the throttle time) otherwise data points might go missing. Also removed the requirement for having data from at least 3 scanners (-1 means any number).