[mod_conference] Avoid a race in member->read_impl access

The conference output loop accesses the member's read_impl while
resampling. The output loop also spawns off the input loop thread, which
is where the member->read_impl is actually initialised. This results in
an uncommon race where the output loop sometimes gets an uninitialised
member->read_impl and bails with logs something like this:

2023-10-12 18:19:01.436844 [DEBUG] conference_loop.c:1340 Setup timer soft success interval: 20  samples: 960 from codec L16
2023-10-12 18:19:01.436844 [DEBUG] conference_loop.c:1497 Outbound conference channel answered, setting CFLAG_ANSWERED
2023-10-12 18:19:01.436844 [NOTICE] switch_core_media.c:15852 Activating write resampler
2023-10-12 18:19:01.436844 [DEBUG] switch_core_media.c:16097 Engaging Write Buffer at 1920 bytes to accommodate 0->1920
2023-10-12 18:19:01.436844 [ERR] switch_core_media.c:16112 Write Buffer 0 bytes Failed!

As a solution, we initialise the member->read_impl even before we start
up the input loop, so that this race can never happen.

src/mod/applications/mod_conference/conference_loop.c

@@ -1316,15 +1316,15 @@ void conference_loop_output(conference_member_t *member)
 	uint32_t flush_len;
 	uint32_t low_count, bytes;
 	call_list_t *call_list, *cp;
-	switch_codec_implementation_t read_impl = { 0 }, real_read_impl = { 0 };
+	switch_codec_implementation_t real_read_impl = { 0 };
 	int sanity;
 
-	switch_core_session_get_read_impl(member->session, &read_impl);
+	switch_core_session_get_read_impl(member->session, &member->read_impl);
 	switch_core_session_get_real_read_impl(member->session, &real_read_impl);
 
 
 	channel = switch_core_session_get_channel(member->session);
-	interval = read_impl.microseconds_per_packet / 1000;
+	interval = member->read_impl.microseconds_per_packet / 1000;
 	samples = switch_samples_per_packet(member->conference->rate, interval);
 	//csamples = samples;
 	tsamples = real_read_impl.samples_per_packet;