KVCacheMode
The cache's serving mode, what every KVLayerSpec.Layout.Auto row resolves to: Continuous (one contiguous run of slots per sequence) or Paged (fixed-size blocks reached through a block table).
The cache's serving mode, what every KVLayerSpec.Layout.Auto row resolves to: Continuous (one contiguous run of slots per sequence) or Paged (fixed-size blocks reached through a block table).