Using ftdm_log_chan() in ftdm_analog_configure_span() is a bad idea,
since the span won't have any channels assigned.

This bug powered by declaring all variables at the top of the function,
even if they are used in an if branch at the end.

A C99'ish:

	if (callwaiting) {
		for (unsigned int i = 1; i <= span->span->chan_count; i++) {
			/* ... */
		}
	}

would have alerted the developer adding the log statement.
But since we can't have nice things (thanks MSVC for not supporting C99!)
</rant>
Signed-off-by: Stefan Knoblich <s.knoblich@axsentis.de>

This partially reverts commit 17d52112.

Some changes in that commit caused 393 files which are in the tree to
be mistakenly ignored.

FS-3121 this is less of a bug and more of a feature request but here you go, that's your quota for the month

FS-3106 --comment-only Try this newer refactored version which completely removes the thread affinity and uses a linked-list to manage the pool, I can run my same 100cps test and only use like 4 db handles

FS-2966 I already fixed this in the wait_for_state, so i copied the same fix to wait_for_state_timeout, the lua one needed a similar call to wait for the state to change to avoid a race, both of you should be fixed with this patch

clear timestamp when generating a fake empty frame to fix edge-case sending the same timestamp over and over

FS-3106 --comment-only Try this out, its got a few elements from your patch but there was a much bigger problem deeper in the code preventing the sqlite handles from being recycled properly

add sip_jitter_buffer_during_bridge which you can set to true to keep a jitter buffer on both ends of the call when you are NormT

add sip_execute_on_image variable similar to execute_on_answer etc so you can run t38_gateway or rxfax etc when you get a T.38 re-invite but no CNG tone or you want to ignore the tone and only react when getting a T.38 re-invite