Fix Sidekiq reporting to Sentry outside of job contexts

We have a module that processes Sidekiq job data before sending it to
Sentry. However, if Sentry is enabled (which means we'll use this class)
and Sidekiq encountered an error outside of a job's context, the
processor itself could break, which is extra-bad: we won't get anything
in Sentry and we actually have two problems.

Sidekiq will still log something like this:

    !!! ERROR HANDLER THREW AN ERROR !!!

But we shouldn't permit that to happen.

The motivating example for this case was having the Sidekiq metrics
server unable to bind to its port because something else was already
using this. By default it uses port 8082 so running this script will
bind to that port and prevent Sidekiq from using it:

    #!/usr/bin/env ruby

    require 'webrick'

    server = WEBrick::HTTPServer.new(BindAddress: "0.0.0.0",
                                     Port: 8082,
                                     DocumentRoot: ".")

    trap("INT") { server.shutdown }

    server.start

Changelog: fixed

lib/gitlab/error_tracking/processor/sidekiq_processor.rb

@@ -53,6 +53,8 @@ module Gitlab
             # 'args' in :job => from default error handler
             job_holder = sidekiq.key?('args') ? sidekiq : sidekiq[:job]
 
+            return event unless job_holder
+
             if job_holder['args']
               job_holder['args'] = filter_arguments(job_holder['args'], job_holder['class']).to_a
             end

spec/lib/gitlab/error_tracking/processor/sidekiq_processor_spec.rb

@@ -178,5 +178,14 @@ RSpec.describe Gitlab::ErrorTracking::Processor::SidekiqProcessor do
         expect(result_hash.dig(:extra, :sidekiq)).to be_nil
       end
     end
+
+    context 'when there is Sidekiq data but no job' do
+      let(:value) { { other: 'foo' } }
+      let(:wrapped_value) { { extra: { sidekiq: value } } }
+
+      it 'does nothing' do
+        expect(result_hash.dig(:extra, :sidekiq)).to eq(value)
+      end
+    end
   end
 end