I had the exact same problem with my python solution. Even tried to use a decorator that I know optimizes recursive function calls. I then rewrote to be non-recursive...same issue with timeouts. This needs to be marked an an issue for python implementation

you are not allocating enough memory, and as a result you are writing out of bounds and corrupting the tests suite:

int *output = malloc(*sz_out);

*sz_out is the length of the array, not its size, because sizeof *output != 1.

You can check out an ASCII table for starters.

How many bytes are moved in the following line of your solution?

  mov cx, [rdi+rsi]   ; char = string[index]

Get a life!

Sorry but I don't see a kata issue there. I am not the final Python translator.