Split Q/K/V inputs with per-head attention results

[jlarson4]

Description

BlockBridge.hook_aliases declared hook_q_input/hook_k_input/hook_v_input/hook_attn_in (pointing at the attention bridge), and AttentionBridge declared these as HookPoint attributes — but nothing fired them. The aliases resolved to a HookPoint, satisfying Tier 1 resolution checks, yet forward bypassed them entirely. Patches at those hooks had zero effect on logits.

This PR makes the hooks live on their intended path. The feature is opt-in via config flags, and when the flags are off the code flows through the existing simple path unchanged. Pure superset, no regression risk for existing callers.

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)

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