Next cursor predictor should limit its output to 4 tokens

[ulugbekna]

[ulugbekna]

verification steps:

1. trigger NES when next edit isn't expected (this will trigger a request to the next-cursor-line predicting model)
2. in the debug log tree, find nes.nextCursorPosition entry which mentions maxResponseTokens to be 4