Remember noise is random. As a matter of probability, it is increasingly rare to get values near -1 or 1, especially with smooth noise such as OpenSimplex.
You have a period of 64, which means "hill" patterns will have that length.
So the general "shape" of the 60x60 area you are testing only represents part of a hill. For that hill in particular it's very unlikely you will reach values near the limits. If we divided space by 64, it would be as if you generated 2x2 numbers to form a square and interpolated corner values inside of it. 4 random numbers is too few to measure.
If you try a larger area, or if you decrease
period, you will notice the minimum and maximum values get closer to -1 and 1.
After testing more, I found that even with 1 octave and a period of 1, the maximum I could get was:
And never beyond. Yet this is over a 10,000 x 1,000 area, and the seed was randomized on each run... maybe there is a flaw somewhere.
The Godot wrapper class looks fine, so this has to do with the C implementation of OpenSimplex itself.