Increase size of CI runner to fix system tests

[krschacht]

A couple days ago we started getting randomly failures like this:

Capybara starting Puma...
* Version 6.4.0 , codename: The Eagle of Durango
* Min threads: 0, max threads: 4
* Listening on http://1[27](https://github.com/AllYourBot/hostedgpt/actions/runs/10258652298/job/28381812667?pr=483#step:9:28).0.0.1:37807
[Screenshot Image]: /home/runner/work/hostedgpt/hostedgpt/tmp/screenshots/failures_test_refreshing_the_page_after_closing_sidebar_keeps_it_closed.png
E

Error:
NavColumnTest#test_refreshing_the_page_after_closing_sidebar_keeps_it_closed:
Net::ReadTimeout: Net::ReadTimeout with "Net::ReadTimeout with #<TCPSocket:(closed)>"
    test/application_system_test_case.rb:20:in `login_as'
    test/system/messages/nav_column_test.rb:6:in `block in <class:NavColumnTest>'

bin/rails test test/system/messages/nav_column_test.rb:86

I tried catching the exception and retrying, but the exception occurs in different parts of the code so that's not easy and probably not the right solution. I tried precompiling assets since some posts suggested this would fix, it didn't help. I spent awhile figuring out how to increase the Capybara timeout since various things online seemed to suggest that part of initializing was taking too long. I finally figured it out with this trail of clues:

• driven_by directive is defined: https://github.com/rails/rails/blob/9ba208c16835f4a174ae9fd385ebc18972d758a4/actionpack/lib/action_dispatch/system_test_case.rb#L158
• the options parameter is passed on through to the driver: https://github.com/rails/rails/blob/main/actionpack/lib/action_dispatch/system_testing/driver.rb#L56
• the options[:timeout] value is read by Capybara: https://github.com/teamcapybara/capybara/blob/0480f90168a40780d1398c75031a255c1819dce8/lib/capybara/selenium/driver.rb#L67
• the PersistentClient inherits from HttpDefault: https://github.com/teamcapybara/capybara/blob/master/lib/capybara/selenium/patches/persistent_client.rb#L5
• the read_timeout: is set from the initializer: https://github.com/SeleniumHQ/selenium/blob/trunk/rb/lib/selenium/webdriver/remote/http/default.rb#L36

But that didn't fix either. Finally, I increased the size of the github action runner just for system tests and that seems to have fixed it. I tried going back to the previous runner but with parallelization turned down to just 1 thread and that still failed.

I'm not sure why this would start happening all of the sudden. Maybe the overall resources that app is using have increased? Maybe github decreased the default (i.e. free) action runner size?

I take that back: the tests ran successfully twice in a row but then started timing out again. In the end, we're using the larger runners (4 cores) but I set parallelization to 2 and I increased the timeout just in case.