Rename models.Bert() to models.BertCustom()

[jbischof]

When prototyping our model API in PR #288 we were following the example of keras-cv. They have DenseNet and DenseNet121 and we have Bert and BertBase. Unfortunately the convention of naming an architecture "base" in NLP makes these names confusing so @mattdangerw and I have discussed changing models.Bert() to models.BertCustom() for clarity.

I think "custom" is a reasonable modifier to indicate that BertCustom is customizable in contrast to BertBase and upcoming arches like BertLarge. Thoughts?