Thank you for this library!
During our internal testing we found an allocation hotspot in Soup.texts. The current implementation is equivalent to running List.append on singleton lists, i.e.

((([s] @ [s]) @ [s]) @ [s]) @ [s] ...

This PR changed Soup.texts to use List.cons and turn the implementation into

s :: s :: s :: ...

This avoids

• allocating the intermediate list (i.e. [s]) then unwrap it
• creating a copy of accumulator list for every List.append in List.fold_left