Desched signal restart can cause behavior difference in tracee

On recent rr/wine, I'm seeing most wine recordings exit with a "partial write" error. It turns out wine is using a writev syscall without proper restart logic here: https://github.com/wine-mirror/wine/blob/8125687441923e11ffb8918b89f3c0d85040b8e2/dlls/ntdll/unix/server.c#L183-L213. Now, this is essentially a wine bug (for which I will submit a patch to wine upstream), but wine doesn't use signals very much, so in the absence of rr it mostly works fine. With rr however, we get a syscall interrupt every time, because this syscall is blocking, so rr's desched signal fires and causes a syscall restart and tracee behavior difference. I'm thinking we should maybe consider automatically restarting syscalls for the tracee if we know that the interrupt was due to a desched signal in order to try to match the untraced behavior as closely as possible.

Start by reading the desched-signal and syscall-restart behavior described in the issue, along with Wine's dlls/ntdll/unix/server.c lines 183-213. Determine whether rr can distinguish desched interruptions from other syscall interrupts and define tests showing that traced Wine behavior matches untraced behavior without breaking restart semantics.