Transformers v5 release: extend xfail condition for TestGRPOTrainer.test_training_vlm_and_liger and update version checks

[qgallouedec]

1. extend xfail condition for TestGRPOTrainer.test_training_vlm_and_liger

Transformers PR

• 🚨 Generalize get_decoder() for multimodal and delete redundant code 🔪  transformers#42156

introduced a Liger-Kernel bug reported in

• AttributeError for 'language_model' in transformers v5 linkedin/Liger-Kernel#960

and fixed in

• Fix missing property access for multimodal models linkedin/Liger-Kernel#966

but not yet released.

TRL currently xfails this test only for dev Transformers (see #4601),
but Transformers 5.0.0 has now been released, so the bug affects stable too. This PR extends the xfail to
Transformers >= 5.0.0 when liger_kernel <= 0.6.4.

2. This PR also replaces transformers 5.0.0.dev0 version check by 5.0.0
3. xfail zero3 test for grpo and aloo, see RLOO and GRPO failing with ZeRO3 #4899

[qgallouedec]

No maintainer is available at the moment, so I will merge this PR without review, so that we can advertise support on day one of transformers v5 with a clean green CI

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[albertvillanova]

Thanks, @qgallouedec!! 🤗