Mirror BERT testing speedups for other models

[mattdangerw]

We can imitate the changes in #859 for all other models in repo.

Let's do this one model at a time, to keep things granular.

You can run a single models test with pytest keras_nlp/models/XX and should see a noticeable improvement in time elapsed if all things go well.

Checklist for a given model:

• Add a serialization test for all classes.
• Mark the saved model testing as "large" for all classes.
• Make the backbone sizes as small as possible while still testing all logic (e.g. num_layers=2).
• Only run a single jit_compile=False test per model class. E.g. test_classifier_fit_no_xla.
• Other minor readability and efficiency improvements (check the original PR carefully).

[mattdangerw]

Also, we are getting a lot of contributor interest right now! Let's all try to..

• Use this issue to communicate.
• Only work on a single model at a time. E.g. wait till you have a PR ready for one model before starting on the next.
• Be patient with each other as we all self organize here :)

Thanks!

[kanpuriyanawab]

Can take this up for Roberta. Thanks.

[TheAthleticCoder]

Can take this up for Debert_v3. Thanks!

[soma2000-lang]

Will take this up for Distill_bert and Albert

[Cyber-Machine]

Will take this issue for f_net model.

[ADITYADAS1999]

I go through bert, thanks

[susnato]

I would like to take this up for xlm_roberta.

[abuelnasr0]

I will take GPT2

[susnato]

I would like to take OPT.

[susnato]

Since no one has taken Whisper and T5 I would I like to take those two.