cleanup: remove unused linear attention helpers

[lightseek-bot]

Summary

• Remove unused migrated helper utilities from linear attention modules
• Drop the unused Mamba state scatter-with-mask kernel and wrapper
• Keep the active GDN, copy/zero, and chunked linear attention paths intact

Tests

• .venv/bin/python -m compileall -q python/tokenspeed/runtime/layers/attention/linear
• git diff --check
• PRE_COMMIT_HOME=/tmp/pre-commit pre-commit run --all-files