soup_message_new_from_encoded_form() claims ownership of the
encoded string, so free it if passed to soup_uri_copy().

The current limit is insufficient for some application services
that require support of at least 20 4K cookie sizes

And release it in the callback. In case of cancellation the item can be
finished by explicitly calling soup_session_process_queue_item()
which can release the last reference before
async_send_request_return_result() is called.

To prepare for the new 'keepalive pong timeout'-feature. See
https://gitlab.gnome.org/GNOME/libsoup/-/merge_requests/389.

GLib 2.69.1 is a development version, so depend on 2.70.0
instead. Also remove all GLIB_CHECK_VERSION checks since
they are not needed any longer.

After this, valgrind reports no leaks any longer when
running websocket-test (with
--suppressions=tests/libsoup.supp).

There is already support for keepalive pings on a regular interval with

    soup_websocket_connection_set_keepalive_interval (...)

Add a new function

    soup_websocket_connection_set_keepalive_pong_timeout (...)

that - if used - begins to require pings to be ponged. If a
ping does not result in a pong before the specified timeout,
the connection is assumed to be lost, and we transition to a
closed state.

To make it somewhat easy to test SoupWebsocketConnection
with valgrind.

Fixes #372

Fixes: https://gitlab.gnome.org/GNOME/libsoup/-/issues/373

Otherwise `--export-dynamic` is added to the linker flags
unconditionally which makes it harder for libsoup3 users to restrict the
exported symbols.

Fixes: 9596a869 ("Detect if libsoup2 is loaded in the same process at init")

Fix wrong user data pointer during request_filter call.

In order to improve debugging add Soup-Host field to ensure that Host
header is always logged.

- The top-level navigation exception only existed for Lax cookies
- Host matching is a direct comparison

Found by Coverity

If the cookie is not well-formed and contains the same attribute twice,
it will leak. Found by Coverity.

I messed up when deciding how to free the GDateTime here, which is not a
GObject. This fixes a regression introduced in !400.

In the past (2015) this was allowed for some problematic sites. However Chromium also does not allow NUL bytes in either header names or values these days. So this should no longer be a problem.