Fix doReadv/doWritev discarding partial count on EAGAIN

[guybedford]

$doReadv reads iovec-by-iovec via FS.read. On a non-blocking socket, FS.read (nodeSockOps.recvmsg) throws EAGAIN when the queue is empty, unlike a regular file which returns 0. If an earlier iovec already consumed data and filled exactly to its length (so the if (curr < len) break doesn't fire), the loop advances to the next iovec, whose FS.read throws EAGAIN. The exception escaped doReadv, discarding the accumulated ret, so the whole readv(2) failed with EAGAIN even though bytes were read.

POSIX readv/writev are single gather operations: they return the available byte count and never fail after partial success.

Observed with an mio TcpStream (NODERAWSOCKETS + PROXY_TO_PTHREAD + JSPI) where an echo returned as two TCP chunks: read_vectored with two iovecs hit recvmsg on iovec #1 (12 bytes, drains recvq) then recvmsg EMPTY on iovec #2 (next chunk not yet arrived), throwing WouldBlock for the whole call. A single-buffer read never hits it — matching "read works, readv fails". It is timing-sensitive: if both chunks are queued before the read, iovec #2 succeeds.

• doReadv and doWritev now catch EAGAIN/EWOULDBLOCK FS.ErrnoError from a subsequent iovec and return the accumulated count when ret > 0, only rethrowing when ret == 0 (nothing transferred yet, so the would-block belongs to the first iovec).

Adds test/fs/test_readv_eagain.c (wired up as test_fs_readv_eagain), which registers custom devices that satisfy the first iovec fully then throw EAGAIN, asserting both readv and writev return the partial count rather than failing. The test aborts without the fix and passes with it.

Made with AI assistance under my review