[worker]: support Gemma 4 per-layer KV cache specs

[Carlos-Marques]

Summary

Adds Gemma 4-aware KV cache spec generation for the Neuron worker.

For multimodal hf_config objects with nested text_config, get_kv_cache_spec() now derives per-layer cache properties from the text config:

layer_types[i] == "sliding_attention" -> head_size=head_dim, sliding_window=sliding_window
layer_types[i] == "full_attention"    -> head_size=global_head_dim, sliding_window=None

This is needed because Gemma 4 E2B mixes local sliding attention (head_dim=256) with full/global attention (global_head_dim=512), so a uniform self.model.head_dim cache spec is incorrect. Unit coverage exercises the nested Gemma4 text config path.

Link to Devin session: https://app.devin.ai/sessions/8501976bc5364801a8d6d8b8b768a547
Requested by: @Carlos-Marques

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring