Add Arcee squared-ReLU MLP adapter (ArceeForCausalLM)

[pablocs116]

Adds TransformerBridge support for Arcee AI's AFM-4.5B (ArceeForCausalLM), a Llama-style dense decoder whose only architectural novelty is an ungated squared-ReLU (ReLU^2) MLP (up_proj -> ReLU^2 -> down_proj) instead of the gated SiLU/GeLU used by Llama.

• Register the "relu2" activation (relu(x)**2) in SUPPORTED_ACTIVATIONS so the config guard and ActivationFunctionFactory accept HF's hidden_act="relu2".
• New ArceeArchitectureAdapter: Llama attention (RoPE, RMSNorm, GQA, no QK-norm, no biases) + ungated MLPBridge. Post-activation ReLU^2 neurons are inspectable via blocks.{i}.mlp.hook_post.
• Register at the four sites (factory, supported_architectures init, model_registry HF_SUPPORTED_ARCHITECTURES + CANONICAL_AUTHORS_BY_ARCH, generate_report descriptions) and add the two checkpoints to the registry.
• Unit + integration parity tests (fp32 + eager); integration asserts bridge logits match HF and hook_post exposes non-negative d_mlp-width activations.

Closes #1467

Description

Please include a summary of the change and which issue is fixed. Please also include relevant motivation and context. List any dependencies that are required for this change.

Fixes # (issue)

Type of change

Please delete options that are not relevant.

• New feature (non-breaking change which adds functionality)

Screenshots

Please attach before and after screenshots of the change if applicable.

Checklist:

• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I have not rewritten tests relating to key interfaces which would affect backward compatibility

[jlarson4]

Looks great thank you @pablocs116