Improve RSPEC parallel task split

Before specs:all rake task did split up all specs on all executors
randomly. This ended up in threads with many small tests idling and
wasting resources.

The new method on the first run assigns tests based on the file
size(more tests means more work) and after the first run buy the
execution time which was measured into the tmp/parallel_runtime_rspec.log file.

Ultimately ensuring a more even distribution of work between the rspec
executors.

Author: FloThinksPi

lib/tasks/spec.rake

@@ -42,9 +42,16 @@ namespace :spec do
   end
 
   def run_specs_parallel(path, env_vars='')
+    group_by_command = if File.exist?('tmp/parallel_runtime_rspec.log')
+                         '--group-by runtime'
+                       else
+                         '--group-by filesize'
+                       end
+
     command = <<~CMD
       #{env_vars} bundle exec parallel_rspec \
       -n `ruby -e 'require "etc"; puts (Etc.nprocessors * (ENV["PARALLEL_TEST_PROCESSORS_MULTIPLE"] || 0.8).to_f).to_i'` \
+      #{group_by_command} \
       --test-options '--order rand' \
       --single spec/integration/ \
       --single spec/acceptance/ \