replication: guard name set fail with try-catch

box_set_instance_name() has a scoped guard to try to restore the
replication back to normal if the connection to master was
recreated to make the applier request a new name.

The guard calls box_restart_replication() which might throw
exceptions. But in this case it apparently can only happen in
theory with bootstrap strategy legacy + high connection quorum +
failing to gather this quorum. Also iostream creation might fail.

Iostream creation failure is not testable, and the legacy
bootstrap strategy is being deprecated at some point. So the
patch simply adds try-catch without a test as it is too much
effort for too little gain to bother. This should also calm down
the coverity reporter.

NO_DOC=bugfix/refactor
NO_CHANGELOG=hardly possible to observe this bug
NO_TEST=too hard to repro