Fix for openvino-tokenizer changes (merge when models are regenerated)

[dkalinowski]

When using older openvino-tokenizer during tokenizer export via convert_tokenizer, tokenizer produces different tokens than in the latest openvino-tokenizer package.

Example (2025.3):
29491, 781, 781, 5, 29560, 7567, 1629,

Now:
29491, 781, 781, 5, 1501, 7567, 1629,

which causes tokenized->detokenized text to miss the space

29560 was [, but 1501 is [