Discord DTLS Buffer Processing

Discord's voice transport may need to assemble ordered DTLS message payloads into fixed-size buffers. Each emitted fragment has a fixed DTLS record overhead, so a buffer can contain fewer payload bytes when it contains more fragments.

Implement pack_dtls_messages(messages, buffer_size, record_overhead). Process messages in order, fragmenting messages when necessary, and return the minimum number of buffers required.

Formal Specification

• messages is a list of non-negative integer payload lengths.
• buffer_size is the maximum number of bytes in every output buffer.
• record_overhead is the number of bytes added for each emitted fragment.
• Return a list of buffers. Each buffer is a list of fragments represented as [message_index, offset, length].
• offset is the payload offset within the original message, and length is the fragment payload length.
• Fragment order must match message order. Empty messages produce no fragments.
• Every fragment consumes record_overhead + length bytes, and each returned buffer must stay within buffer_size.

Use a greedy strategy: fill the current buffer with the largest possible next fragment. Start a new buffer when the next fragment cannot fit.