Add support for summarization of longer texts.

[michalharakal]

Usually are summarization models trained by text blocks with limited size, e.g. 800 characters for mt5-small-sum-de-en-v1.
To achieve the best prediction performance, also the input text has to be limited to this length.

As one of solution for summarizing of longer texts, following approach could be tested:

• split input text for prediction into blocks on sentences level, where blocks character count is limited
• predict a summary for individual block
• join results of blocks summary into single text block.