LIDAR is probably the most accurate 3D scanning solution we have, but even it is not typically millimeter accurate. Often times this accuracy is scan distance dependent and we may be able to scan smaller objects with much higher accuracy at shorter distances, while achieving lesser accuracy at longer throws. We can derive a lot of information about something using survey data, but that survey data goes far beyond just what the raw LIDAR scan is capturing, and typically when processing that data we will want to calibrate the entire point cloud with known accurate measurements taken on set. It is not uncommon to drop a ruler next to things you're trying to 3D scan, to ensure that the ruler itself gets scanned with the object and can be used to calibrate the dataset.
Additionally, high quality LIDAR solutions can be incredibly expensive. One popular such LIDAR solution used in the visual effects industry is the Leica BLK360 which is 16,000$, which is ironically one of the cheaper solutions in its category and is intended for industrial scanning solutions. The BLK360 has an advertised accuracy of 4mm at a distance of 10m, which is quite good. This is the kind of scanner that i believe was used to historically preserve the Notre Dame cathedral and assist with its reconstruction.
Some LIDAR scanners like the ones made by Artec (60,000$+) advertise sub-millimeter accuracy but i am skeptical of these claims as almost all 3D scans will require some kind of cleanup in post.
For example, 3D scan store uses a different solution to LIDAR, called photogrammetry, because their interest is in capturing high quality texture detail which is not typically achievable with LIDAR which normally only captures spatial details. I know that they utilize the Ten24 capture studio which likely cost millions of dollars to design, and yet, here is a breakdown of their scan cleanup process:
https://www.youtube.com/watch?v=qZn69bVyT0o&ab_channel=3DScanStore
You can see that even though their scans are very high fidelity, and i would say they are probably some of if not THE best in the industry at what they do, there is still artefacting and noise that must be cleaned up and sculpted over by hand by a skilled technician / sculptor.
Photogrammetry is an entirely different process than LIDAR and is used by calculating the difference between different camera angles and features, or points of high contrast in color or luminance, to triangulate and solve for the positions of each relative to each other. It is probably also the cheapest solution to 3D scanning that exists, especially for doing high quality scans of small objects, and like LIDAR it is assisted by calibrating for known variables like the camera focal length, the sensor size, the known real world dimensions of objects in the scene, these can all give it better results. But it has its own set of problems, like being wholly unable to deal with highly transparent or reflective surfaces. 3D Scan Store, for instance, specializes in digital doubles of humans, so they use a cross polarizing filter to remove all reflections from the surface of the skin and attempt to evenly light the human from all angles to naturally and optically extract albedo. Additionally, natural micro tremors in any human being will ultimately throw off the quality of the scan, so it must be captured from all angles at the same time. They have massive rigs with hundreds of high quality DSLR cameras, linked up to server farms designed to trigger them all at once and then dump massive amounts of raw photos all at once which is why their professional level rigs are so expensive while amateur scanning of smaller objects can sometimes be as easy as setting up a turntable with a softbox in front of a white screen with a single camera. Bit depth, resolution, sharpness, exposure, surface detail and lighting quality all play a role in getting a higher quality scan, it is not as simple as specifying mm or cm level accuracy.
LIDAR does not suffer from the same issue with reflective materials, but is ultimately far more expensive to do right. Its actually quite impressive that Apple was even able to fit a lidar scanner into their iphones for the price they did, and the quality is quite impressive but nowhere near professional quality and ultimately skillfully shot and processed photogrammetry will get you much further. It is more useful to have this level of fidelity LIDAR for things like motion and performance capture, but that is another topic entirely.
The good news is that with some effort LIDAR data can be combined with photogrammetry data in some applications and studios will often use a combination of techniques to pull off their on set surveys. I actually saw a breakdown from the avengers where they combined their LIDAR scans of entire city blocks in NYC with HDRI probes and then reprojected the 360 images onto the resulting LIDAR geometry with positional accuracy. They probably only had a brief moment to lock down those city blocks and shoot their survey data and that was the quickest way to do it, typical photogrammetry would have taken too long.
Hacked microsoft kinects were used by homebrew VFX guys back in the day (usually as garage experiments more than professional use cases) using a different technology to LIDAR called structured light. Instead of laser pulses, it uses infrared dots or lines projected in a pattern across the subject and then uses infrared cameras to detect the distortions of those lines. A variation on this technology uses stereo infrared cameras to also try to do a bit of triangulation and hopefully improve the quality of the solver. Microsoft later released the kinect azure aimed at industrial applications rather than gamers, but i don't know anyone who still uses them.
Most depth sensors on phones are not LIDAR at all, but the much lower quality TOF or stereo sensors. TOF works similarly to LIDAR but is much less accurate, really only good for pasting a snapchat filter onto a tracked face, often with issues even properly tracking faces. Intel sells a wide variety of sub-1000$ realsense sensors using LIDAR, TOF, or stereo sensors but from experience i don't recommend any of them as they are... quite shit. Unusably shit.
Hopefully this gives you an idea of the plethora of 3D scanning solutions that exist out there and how varied their use cases are. Probably the only thing that the average consumer is going to be able to afford is iPhone LIDAR or photogrammetry. But its always getting better and cheaper every year.
