It sounds like there are several questions here, I will try to the best of my ability to answer them.
"However, today my Garmin has especially bad day. After it took 15 minutes to get GPS signal, it blocked at about 170 bpm for the whole rest of the run. I could have much at the moment of the GPS lock, because it was a slope, but later, it was rather a mild run."
This could have a few different reasons. Perhaps it was simply a bug/inaccurate reading in the Garmin, but more likely it was cardiovascular drift. This refers to the natural increase in heart rate that occurs when running with no change in pace. It is mostly caused by the natural increase in core body temperature, similar to how running in hot weather will cause higher heart rates. What likely happened is your increased body temperature was increasing your heart rate at roughly the same rate that your slower pace was causing your heart rate to slow down.
"I've observed on the occasion, that Garmin tends to make up data when it has problem getting them. For example, when I SUP and the watch is on the top of the suit, it displays the heart frequency I've got when I took it off the hand, and extrapolates it for the whole lenght of the activity."
I don't own a Garmin smart watch and do not know why it would extrapolate data for the length of the activity, but I do know that it will not work properly on top of a wet suit. Different brands will use different technologies to measure HR, see How do fitness trackers measure your heart rate? In short some (Including Garmin) use lights that flash at your wrists and measure the feedback. If you block these lights with something like a wet suit, it will not work. Many smart watch manufacturers specify that it must be snugly fit 2 finger widths above the wrist bone in contact with skin for the HR feature to work. Garmin's support misses the skin part, but I believe it was just an oversight. Garmin's support on how to wear the watch
How much can you trust data from smart watches?
TLDR - Wrist worn HR monitors vary by brand but are largely accurate enough for the average person for most scenarios. The more intense the exercise, the less accurate they get. Even at higher intensities, they should still be accurate enough for the vast majority of people. If you demand perfect accuracy, a chest HR monitor might be better, but they are much less convenient.
One issue with smart watches is that we wear watches on our wrists and the wrist is full of bones, tendons, and muscles that create optical noise in the wrist making it very difficult to measure heart rate accurately. See Heart Rate Monitor Location Matters. Notably in this article, they missed one of the better locations, the chest.
Because wearable fitness items are so popular, there are tons of studies on this. I am only going to briefly go over two. The first study specifically focuses on high levels of exertion, which I believe is more suitable for you. The second one looks into how accurate they are in general activities.
Accuracy of commercially available heart rate monitors in athletes: a prospective study
This study included subjects that were 18+ years old, could run a mile in 7 min or less, did not have tattoos around their wrists, and no known health issues that could impact the results. They attached a chest HR monitor as something to measure the watches against, since that was the "gold standard for HR measurement". Participants also wore different wrist HR monitors on their wrists. (Note: previous studies demonstrate that in healthy individuals there is no difference in accuracy of HR monitors based on which wrist they are worn on.) They then had the subjects run at different speeds and compared the values on the wrist monitors to the chest. Each watch was tested 25 times. Something they noted is that the more intense the exercise the more inaccurate the results from the wrist monitors. The results of the study confirmed that wrist devices were not as accurate as the chest strap, but the Apple Watch III and the Fitbit Iconic came the closest to the ECG standard.
Their results show the concordance correlation coefficient. The closer to 1, the more accurate that device is.
- Polar chest strap .98
- Apple Watch III .96
- Fitbit Iconic .89
- Garmin Vivosmart HR .89
- Tom Tom Spark 3 .89
Wrist-worn optical and chest strap heart rate comparison in a heterogeneous sample of healthy individuals and in coronary artery disease patients
In this study, subjects used a wrist monitor and compared it to a ECG via chest strap in a broad range of activities. They concluded that the mean absolute error was not larger than 3 bpm, which they say is "clinically acceptable for a number of applications."