feat: add socket buffer size configuration

[wemeetagain]

Looking at performance here it appears quic has significantly reduced throughput compared with tcp. So what gives?
It may be that the OS buffer size is a factor. Whereas tcp typically has a growable OS-managed buffer, udp's buffers are fixed unless otherwise set by the application (or set via OS-wide configuration).

This PR exposes the send and receive buffer sizes (SO_SNDBUF and SO_RCVBUF) to be configurable.
Note that the OS imposes its own limits, so any values larger than the OS limits (typically for most OSs) will get clamped to the limit.

Various (semi)related links:

• Consider increasing OS socket buffer (SO_SNDBUF, SO_RCVBUF) mozilla/neqo#1962
• https://yomo.run/docs/devops_tuning

TODO figure out good defaults

[achingbrain]

LGTM