Fix in the example in examples/retrieval

[hectorem2]

Quick fix in the retrieval example so that only at most n_seq_max chunks of
text are added to the llama_batch and this example stop crashing due attempting
to process more token sequences in a batch than allocated by llama_context::
encode and llama_context::decode.

The retrieval example crashes because (according to my understanding of the llama_context code) llama_context::encode is going to allocate n_seq_max sequences of tokens, and the loop in the example keeps adding more sequences. I added a condition so that the loop stops adding tokens when s reaches llama_n_seq_max(ctx).