add 3 more small models

[mike-ferguson]

Add TinyLlama 1.1B, Pythia 2.8B, and OPT 2.7B model plugins

Changes

• TinyLlama 1.1B (brainscore_language/models/tinyllama1_1b/): Registers TinyLlama/TinyLlama-1.1B-Chat-v1.0 via HuggingfaceSubject, mapped to model.layers.21 (last of 22 layers, hidden size 2048).
• Pythia 2.8B (brainscore_language/models/pythia2_8b/): Registers EleutherAI/pythia-2.8b via HuggingfaceSubject, mapped to gpt_neox.layers.31 (last of 32 layers, hidden size 2560).
• OPT 2.7B (brainscore_language/models/opt2_7b/): Registers facebook/opt-2.7b via HuggingfaceSubject, mapped to model.decoder.layers.31 (last of 32 layers, hidden size 2560).
• Modifies the test_identifier test in phi-2 to be correct.

Each plugin includes tests for model loading, identifier, neural representations and next-word prediction.

Notes

• All three are open (non-gated) models — no HuggingFace token required.
• Layer mappings are defaults; can be updated after benchmark-driven selection.

[mike-ferguson]

tests are failing due to OOM, to be expected. Manual merge