There is a difference between the roulette example and the others.
In the roulette example, there are not enough samples to figure out whether the results were a coincidence or a real trend. In stats you can actually calculate the chance the wheel or coin is biased. And when you get to around 1,000 flips, even so much as a 600 heads will indicate certainty of bias, and a ridiculously low 0.00something% chance it was due to coincidence.
But the chart you showed of the earth cooling isn’t just random noise with streaks like a person playing roulette; it has enough data points to show that it’s a real trend of the earth cooling over millions of years.
Similarly with the recent data we have enough data points to know for sure (or if you prefer 99.99something% sure) that the recent warming of the Earth over the past several decades coincides with human carbon output. It’s especially obvious if you zoom in on the past 10,000 or so years.
“If you buy a stock because the overall trend of its graph is going up, you will learn the hard way by losing lots of money.” This is in fact not the case. Even someone who bought right before the 2007 recession would be at a net positive now. Also the crashes have identifiable causes like human or computer behavior. I don’t think Earth has a bunch of Wall Street climate-manipulators ready to crash the temperature just yet.