Its not about being far away from previous number, but the number of roll since the last occurrence of the number.
If there is 50 rolls without a one, you can still have more chance winning playing one in the next rolls.
But it will rarely reach even 50 rolls, and on the large number of roll you will still come back to average.
That's incorrect. You have the same chance to roll a 1 at any point during the game regardless of what (or how often) was rolled or not rolled before, otherwise the RNG would be flawed. There is no purposeful coming back to average. The average is the consequence of a good RNG, not something the RNG tries to simulate.
In other words, if you hit an "unusual" streak of below-expected-average numbers the RNG will not generate above-expected-average numbers to compensate. The actual average will get closer to the expected average in a (very) long run as your "unusual" streak will have less and less weight in the total.
If its a computer RNG yes, if Its about statstics on random occurrence of an event in nature, and fair gambling game then no. Depend what is the purpose of the RNG its not the same if Its To generate secure password/key, for simulating natural pattern like fractal or perlin, if Its for gambling etc
But its the purpose of the algorithm that i posted before to make sure it tends toward a mean in the long run. How long is the run being dependant of the number of possibles values.