Refactor paged attention dispatch to support multiple attention types

[WindChimeRan]

Summary

• Extract attention-type-specific logic from the monolithic wrapper into per-type modules (attention_sdpa.py, attention_linear.py)
• The wrapper now dispatches based on module attributes: paged varlen SDPA path for standard dot-product attention (MHA/GQA/MQA), stub for linear attention (GatedDeltaNet), and this will also unblock MLA for glm-4.7-flash and deepseek.
• Change layer patching from single-attribute (self_attn on all layers) to per-layer lookup — required for hybrid models like Qwen3.5 where some layers use self_attn and others use linear_attn
• Add xfail integration test for Qwen/Qwen3.5-0.8B with paged attention

No new features. This is a refactor to unblock collaboration on Qwen3.5 linear attention support.

Why attribute-based detection?

We wrap mlx_lm/mlx_vlm attention modules at runtime without modifying their source. Since we don't own the model code, we detect attention type by probing module attributes (e.g. q_proj + o_proj → SDPA, conv1d + no q_proj → linear). This should works across all known GatedDeltaNet variants (qwen3.5)

How to add a new attention type

1. Create attention_<type>.py with is_<type>() detector + <type>_forward() implementation
2. Add one elif branch in the wrapper dispatch (paged_attention.py)

Test plan

• Qwen3.5-0.8B paged attention xfail hits NotImplementedError on linear attention layers
• Passed the new deterministic test from Update deterministic test golden tokens for MLX 0.31 #202. The refactoring doesn't break the existing paged varlen kernel on Qwen3-0.6B

[ericcurtin]

Minor unused import in paged_attention.py, no unit tests for dispatch logic (only slow integration test)

[WindChimeRan]

Thanks for the review! @ericcurtin

Added 4 fast tests (test_attention_dispatch.py) that verify the detection heuristics against real mlx_lm modules. qwen3.Attention, qwen3_5.DecoderLayer (both SDPA and GatedDeltaNet layers), and qwen3.Model for find_layers. No model weights needed, runs in ~2s. The full dispatch path (wrapper → forward → Metal kernel) is covered by the existing integration test.