ollamarunner: Use a separate context per multimodal input

Currently there is a single context per sequence, shared all by
all multimodal inputs. Since we build a vision encoder graph per
image, with a large number of inputs we can eventually hit the
maximum number of graph nodes per context.

This changes to use a separate context for each image, ensuring
that available resource limits are consistent.

model/models/gemma3/model.go

@@ -111,7 +111,7 @@ func (m *Model) EncodeMultimodal(ctx ml.Context, multimodalData []byte) (any, er
 	return visionOutputs, nil
 }
 
-func (m *Model) PostTokenize(ctx ml.Context, inputs []input.Input) ([]input.Input, error) {
+func (m *Model) PostTokenize(inputs []input.Input) ([]input.Input, error) {
 	var result []input.Input
 
 	for _, inp := range inputs {