Universal Sentence Encoder for Russian based on RuModernBERT, with support for context lengths up to 8,192 tokens and Matryoshka representation lear