When reading from the socket with `sock_tcp_read()` it would only return
data from at most one internal connection buffer, even if the buffer
passed to `sock_tcp_read()` is larger and there is more data available
in the connection. This patch makes `sock_tcp_read` process all the
available data so long as there's more data to read available
immediately.
lwip receives network buffers that are made available to lwip_sock_tcp
calling `netconn_recv_tcp_pbuf`. The size of these buffers depends on
the size of the network packets. An application calling `sock_tcp_read`
can pass any arbitrary buffer size to read (copy) from these internal
network buffers, which may be smaller. lwip_sock_tcp keeps around the
last network buffer (`struct pbuf last_buf`) and the offset into this
buffer already consumed by the application (`last_offset`).
However, when multiple application reads from the same `pbuf` buffer
occur, the `last_offset` must be updated incrementing it. The code had
a bug that would work only when `last_offset` was either 0 (no previous
partial read) or when the current read was consuming all the remaining
data (when `buf_len == copylen`).
This patch fixes the issue an allows multiple reads from the same
buffer.
Either the sock is provided with `sock_*_send()` or not. In the first
case the indirection is not necessary, and in the second we need to
delete the created `conn` within `lwip_sock_send()` anyway, so returning
it makes no sense.
