Fix: bt2: clear Python error indicator in trace and trace class destruction listeners

Fix: bt2: clear Python error indicator in trace and trace class destruction listeners

When an exception is raised in a trace or trace class destruction
listener, we hit this assert:

    LIB/TRACE-CLASS destroy_trace_class@trace-class.c:110 Babeltrace 2 library postcondition not satisfied; error is:
    LIB/TRACE-CLASS destroy_trace_class@trace-class.c:110 Destruction listener kept a reference to the trace class being destroyed: tc-addr=0x6080000027a0, tc-is-frozen=0, tc-stream-class-count=0, tc-assigns-auto-sc-id=1
    LIB/TRACE-CLASS destroy_trace_class@trace-class.c:110 Aborting...

The problem is that we don't clear the error indicator on the error code
path after running the user code.  The error indicator (the exception)
keeps a reference to the Python stack, which has a reference to the
trace or trace class Python object.  This, in turn has kept a reference
to the Babeltrace trace or trace class object.

The solution is to clear the error indicator, which releases all of
this.

Another problem is that calling loge_exception_append_cause appends an
error cause, even though it's not valid to leave an error cause when
exiting those functions, as they don't return an error status.  That
error cause is left dangling and might wrongfully appear on an eventual
future error stack.

To fix it, use logw_exception instead of loge_exception_append_cause.
This matches what we do in component_class_finalize and
component_class_message_iterator_finalize, which are very similar
situations.

Change-Id: Ib756461610366badb6384577ad7cdf6468944be9
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/2224
Reviewed-by: Francis Deslauriers <francis.deslauriers@efficios.com>