Split AD tests into workflows

[Red-Portal]

This PR separates the AD backend integration tests into their own workflow. This has the following benefit:

• The test codes become simpler since we don't have to loop over the various AD backends
• The test results can be viewed at a glance from the GitHub Actions page
• We can monitor the AD backend support via GitHub badges.

All AD integration tests invoke .github/workflows/AutoDiffIntegration.yml, which implements GitHub's reusable workflows feature.

[github-actions]

AdvancedVI.jl documentation for PR #190 is available at:
https://TuringLang.github.io/AdvancedVI.jl/previews/PR190/

[github-actions]

Benchmark Results

Benchmark suite	Current: 5928201	Previous: e801817	Ratio
normal/RepGradELBO + STL/meanfield/Zygote	3869106797 ns	4047146731.5 ns	0.96
normal/RepGradELBO + STL/meanfield/ReverseDiff	1128292103 ns	1118748915 ns	1.01
normal/RepGradELBO + STL/meanfield/Mooncake	1190705399 ns	1241121652 ns	0.96
normal/RepGradELBO + STL/fullrank/Zygote	3955271108 ns	4002369352.5 ns	0.99
normal/RepGradELBO + STL/fullrank/ReverseDiff	1676757225 ns	1619611799 ns	1.04
normal/RepGradELBO + STL/fullrank/Mooncake	1281361888 ns	1285127945.5 ns	1.00
normal/RepGradELBO/meanfield/Zygote	2767657958 ns	2834638479.5 ns	0.98
normal/RepGradELBO/meanfield/ReverseDiff	788761749 ns	788049327 ns	1.00
normal/RepGradELBO/meanfield/Mooncake	1062465846 ns	1093464215 ns	0.97
normal/RepGradELBO/fullrank/Zygote	2785921421.5 ns	2873772655.5 ns	0.97
normal/RepGradELBO/fullrank/ReverseDiff	972573810 ns	956640380.5 ns	1.02
normal/RepGradELBO/fullrank/Mooncake	1118695995 ns	1161037456 ns	0.96
normal + bijector/RepGradELBO + STL/meanfield/Zygote	5484603213 ns	5705749184 ns	0.96
normal + bijector/RepGradELBO + STL/meanfield/ReverseDiff	2380715424 ns	2472318329 ns	0.96
normal + bijector/RepGradELBO + STL/meanfield/Mooncake	4098330520.5 ns	4249849133.5 ns	0.96
normal + bijector/RepGradELBO + STL/fullrank/Zygote	5615396878 ns	5732121704 ns	0.98
normal + bijector/RepGradELBO + STL/fullrank/ReverseDiff	3047981874.5 ns	3070250822 ns	0.99
normal + bijector/RepGradELBO + STL/fullrank/Mooncake	4210618967 ns	4377541897 ns	0.96
normal + bijector/RepGradELBO/meanfield/Zygote	4309224236 ns	4461833085.5 ns	0.97
normal + bijector/RepGradELBO/meanfield/ReverseDiff	2030629168 ns	2088736319 ns	0.97
normal + bijector/RepGradELBO/meanfield/Mooncake	3930307061 ns	4122678327.5 ns	0.95
normal + bijector/RepGradELBO/fullrank/Zygote	4353748197 ns	4571360982 ns	0.95
normal + bijector/RepGradELBO/fullrank/ReverseDiff	2295184083 ns	2345239376 ns	0.98
normal + bijector/RepGradELBO/fullrank/Mooncake	4071153970.5 ns	4217237241 ns	0.97

This comment was automatically generated by workflow using github-action-benchmark.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@981bcc7). Learn more about missing BASE report.
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #190   +/-   ##
=======================================
  Coverage        ?   88.98%           
=======================================
  Files           ?       18           
  Lines           ?      445           
  Branches        ?        0           
=======================================
  Hits            ?      396           
  Misses          ?       49           
  Partials        ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[penelopeysm]

I think this setup feels a bit complex. Could you just have two different tests: those which use AD and those which don't? In other words:

• turn 'General' into 'AD' and maybe just rename it ReverseDiff.yml
• group the families and paramspacesgd tests together

[Red-Portal]

Yeah I agree. I actually made things simpler.

[Red-Portal]

I wonder if we should go back to ReTest since it seems to be a little better maintained than before. The test filtering feature there is really useful.

[penelopeysm]

I'm quite agnostic towards frameworks like ReTest. If you think it's useful I wouldn't be opposed to it. I've been trying to use https://github.com/theogf/TestPicker.jl recently which (I think) does similar things too with test filtering

[penelopeysm]

Looks good :)