This seems to be a strange one, reported by djbog (Scott Marley, @s-marley ) in the reddit group. I ran my own test (on an UNO) and got the same results.
When running inoise8(i) it returns a constant value of 128 from the range of i = 3061 to 3599.
for(uint16_t i = 0; i < 5000; i++) {
Serial.print(i);
Serial.print(", \t");
Serial.println(inoise8(i));
}
Here is Scott's reddit post with a plot that shows the "flat" in the otherwise nice curvey output.
https://www.reddit.com/r/FastLED/comments/ju7ams/something_i_dont_understand_with_inoise8x_output/
Any ideas why it would do this?

Hrm. Interesting. Thank you for the issue report here and the _really_ useful graph.
I'll assign this to myself and queue it up to take a look. I suspect it's a loss-of-precision issue somewhere in the math.
Do we know if both examples were run on the same processor/architecture? It would be interesting to test on a couple different ones to see if it makes any difference. I should have a M4 and esp32 I can try it on. I'll report back.

Looks to be the same on ESP32 using arduino 1.8.13 and FastLED 3.3.3
Most helpful comment
Hrm. Interesting. Thank you for the issue report here and the _really_ useful graph.
I'll assign this to myself and queue it up to take a look. I suspect it's a loss-of-precision issue somewhere in the math.