ml: Allow models to constrain inputs to a single batch

Models may require that a set of inputs all be processed as part
of the same batch. For example, if an image has multiple patches
with fully connected attention between them, we should not split
the batch in the middle of an image.

Fixes #9697

model/input/input.go

@@ -15,6 +15,12 @@ type Input struct {
 	// stored in Multimodal, used for caching and comparing
 	// equality.
 	MultimodalHash uint64
+
+	// SameBatch forces the following number of tokens to be processed
+	// in a single batch, breaking and extending batches as needed.
+	// Useful for things like images that must be processed in one
+	// shot.
+	SameBatch int
 }
 
 // MultimodalIndex is a multimodal element (such as an image)