mlxrunner: Fix panic on full KV cache hit

When the entire prompt was already cached (e.g. repeated prompt),
findRemaining returned an empty slice, causing FromValues to panic
on an index-out-of-range accessing a zero-length byte slice.

Fix by always keeping at least one token to re-evaluate so the
pipeline can seed token generation. Also reject empty prompts
early rather than panicking.

x/mlxrunner/cache.go

@@ -78,6 +78,12 @@ func (c *kvCache) findRemaining(tokens []int32) []int32 {
 		prefix++
 	}
 
+	// Always keep at least one token to re-evaluate so the
+	// pipeline can seed token generation from it.
+	if prefix == len(tokens) && prefix > 0 {
+		prefix--
+	}
+
 	if prefix < len(c.tokens) {
 		trim := len(c.tokens) - prefix
 		for _, kv := range c.caches {

x/mlxrunner/pipeline.go

@@ -47,6 +47,10 @@ func (r *Runner) TextGenerationPipeline(request Request) error {
 	mlx.ResetPeakMemory()
 
 	inputs := r.Tokenizer.Encode(request.Prompt, true)
+	if len(inputs) == 0 {
+		return errors.New("empty prompt")
+	}
+
 	session := r.cache.begin(r.Model, inputs)
 	defer session.close()
 
@@ -54,7 +58,6 @@ func (r *Runner) TextGenerationPipeline(request Request) error {
 	tokens := session.remaining
 
 	total, processed := len(tokens), 0
-	slog.Info("Prompt processing progress", "processed", processed, "total", total)
 	for total-processed > 1 {
 		if err := request.Ctx.Err(); err != nil {
 			return err
@@ -104,7 +107,6 @@ func (r *Runner) TextGenerationPipeline(request Request) error {
 		nextSample, nextLogprobs = step(sample)
 
 		if i == 0 {
-			slog.Info("Prompt processing progress", "processed", total, "total", total)
 			mlx.Eval(sample)
 			final.PromptTokensDuration = time.Since(now)
 			now = time.Now()