websocket: support CURLOPT_READFUNCTION

[icing]

Add support for CURLOPT_READFUNCTION with WebSocket urls when not in connect-only mode, e.g. when using curl_multi_perform.

Install the callback function and set CURLOPT_UPLOAD. Return CURL_READFUNC_PAUSE when having nothing more to send and unpause the transfer when more data is ready.

This will send the read bytes in a WebSocket BINARY frame. For other frame types, add a new API function:

/*
 * NAME curl_ws_start_frame()
 *
 * DESCRIPTION
 *
 * Buffers a websocket frame header with the given flags and length.
 * Errors when a previous frame is not complete, e.g. not all its
 * payload has been added.
 */
CURL_EXTERN CURLcode curl_ws_start_frame(CURL *curl,
                                         unsigned int flags,
                                         curl_off_t frame_len);

Add an example using CURLOPT_READFUNCTION and curl_ws_start_frame() in docs/examples/websocket-updown.c.

Add support for this mode in the pytest "ws_data" client and have all tests run in 'curl_ws_send/recv' and 'perform' mode as well.

This allows on the command line to "talk" to an echo server:

> curl --no-progress-meter  -T . wss://echo.websocket.org

so stdin and stdout are the websocket connection with auto-en- and decoding. For libcurl users, they can multi handle many such websocket transfers and have READFUNCTION and WRITEFUNCTION to do with the data whatever they like.

Update: when CURLWS_RAW_MODE is set for the transfer, the websocket encoding client reader is not installed and the CURLOPT_READFUNCTION can add raw websocket frames.

[bagder]

This branch has conflicts that must be resolved

[bagder]

this branch has conflicts ...

[icing]

I wonder if we should add a --ws-raw option to the command line, for when people want to use a wss:// pipe and manage the websocket protocol frames themselves. It would only work for a single url, but might be handy.

Update: on second thought, let's do this if someone really needs it.