runner: clear cache when shift is not possible (#9433)

Clear KV cache when shift operation is not supported by model.
Added KvCacheCanShift() check to handle models that can't perform cache shifts,
falling back to full cache clear while preserving logical token history to
maintain expected behavior when context window fills up.

runner/llamarunner/runner.go

@@ -389,7 +389,15 @@ func (s *Server) processBatch(tokenBatch *llama.Batch, embedBatch *llama.Batch)
 				if len(seq.pendingInputs) == 0 {
 					err := s.cache.ShiftCacheSlot(seq.cache, seq.numKeep)
 					if err != nil {
-						return err
+						var reprocess *ErrReprocessInputs
+						if errors.As(err, &reprocess) {
+							// Prepend these inputs to the sequence's inputs queue for reprocessing
+							seq.inputs = append(reprocess.Inputs, seq.inputs...)
+							// Continue processing as normal
+							continue
+						} else {
+							return err
+						}
 					}
 				} else {
 					break