fsm: Change semantics of LOGPFSML() log-level

The general idea about each osmo_fsm_instance having a separate
log-level was to be able to selectively increase/show/enable logging
for some FSM instances (e.g. of a particular subscriber) while
maintaining normal logging verbosity for all other instances of the
same FSM.

The introduction of LOGPFSML() in Change-Id
If295fdabb3f31a0fd9490d1e0df57794c75ae547 broke that idea, as it would
use a compile-time log level, irrespective of the
osmo_fsm_inst.log_level setting of the given instance.

Let's combine the two:
Use the explicit level stated at LOGPFSML(), _unless_ this instance
has a higher log_level configured.

This way, all FSMs should normally be created with
osmo_fsm_inst.log_level == LOGL_DEBUG.  At that point LOGPFSM()
statements would be rendered at debug level, typically below the
threshold of most logging configurations.

Code that has explicit higher log levels like LOGPFSML(fi, LOGL_ERROR)
would always be printed, as it is an error message.

And if we now increase the osmo_fsm_inst.log_level, then even the normal
LOGPFSM() statements would suddenly be logged at that higher level,
selectively increasing log verbosity - like originally intended.

Change-Id: I1820f04d0c6f5d5ff08eb95b8c0e88764534491a

include/osmocom/core/fsm.h

@@ -120,7 +120,8 @@ struct osmo_fsm_inst {
 void osmo_fsm_log_addr(bool log_addr);
 
 #define LOGPFSML(fi, level, fmt, args...) \
-		LOGP((fi)->fsm->log_subsys, level, "%s{%s}: " fmt, \
+		LOGP((fi)->fsm->log_subsys, OSMO_MAX(level, (fi)->log_level), \
+			"%s{%s}: " fmt, \
 			osmo_fsm_inst_name(fi),				    \
 			osmo_fsm_state_name((fi)->fsm, (fi)->state), ## args)