bugfix: silent dataloss in udp socket send_to_many call (sendmmsg).

[lvboudre]

Problem

The current implementation of send_many_to calls packets.clear() regardless of the return value of the sendmmsg syscall. On Linux, sendmmsg is not atomic; it iterates through the provided headers and may return early if the socket's send buffer (SO_SNDBUF) becomes full or if the kernel encounters backpressure.

When this happens, sendmmsg returns the number of messages successfully queued (which may be less than the total requested). By calling .clear(), we were silently dropping any packets that the kernel had not yet accepted.

Solution

this PR makes sure we loop over all remaining packet until they are all succesfully sent to the kernel.

---

This change is 

[lvboudre]

I deleted unit test that I have written, for some reason they don't work on your runner :/ .
In order to create the bug you have to send alot of data really fast, so a little bit hard to reproduce in unit test.

[dlon]

Nice find! Looks good to me. We're discussing licenses right now, which prevents us from accepting outside contributions, but we'll get back to you asap.