NaN rotaty_emb weights in exported pretrained models

I used `pretraining_documentation/convert_to_hf.py` to convert pretrain checkpoint pt to HF models, as some model variants don't need to extend context window, but I found that output logits are nan, and `rotaty_emb` is nan in local attention layers

The fix is simple, just change `config.json` content:

* old: `local_rope_theta: -1`
* fixed: `local_rope_theta: 10000.0`, the same as global and as in training

and reload model from dir again, then model outputs totally make sense.

I guess this is because pretrain yamls don't have this value, and FlexBERTConfig [defaults to -1](https://github.com/AnswerDotAI/ModernBERT/blob/8c57a0f01c12c4953ead53d398a36f81a4ba9e38/src/bert_layers/configuration_bert.py#L94), but I'm not quite sure what would be a proper fix.

Start with pretraining_documentation/convert_to_hf.py and inspect how the exported config.json gets its local_rope_theta value. Compare that path with FlexBERTConfig in src/bert_layers/configuration_bert.py and the pretraining YAMLs; done means converted local attention rotary embeddings and model logits are no longer NaN without manually editing config.json.

Written by the indexing model from the issue text.