consolidate the tokenizer (#14327)

This change adds a new x/tokenizer package which includes:
  * New BPE and SentencePiece tokenizers
  * Removing the dependency on the imagegen tokenizers
  * Fixes to multibyte decoding in the pipeline
  * Various correctness and benchmark tests

Not included in this PR is the WordPiece tokenizer for BERT models which will be
added when we add embedding models. The imagegen tokenizers will also be removed in
a follow-up PR.

x/mlxrunner/utf8_buffer_test.go

@@ -0,0 +1,46 @@
+package mlxrunner
+
+import (
+	"bytes"
+	"testing"
+)
+
+func TestFlushValidUTF8Prefix_PreservesIncompleteRune(t *testing.T) {
+	var b bytes.Buffer
+
+	b.Write([]byte{0xE3, 0x81})
+	if got := flushValidUTF8Prefix(&b); got != "" {
+		t.Fatalf("first flush = %q, want empty", got)
+	}
+
+	b.Write([]byte{0x93, 0xE3})
+	if got := flushValidUTF8Prefix(&b); got != "こ" {
+		t.Fatalf("second flush = %q, want %q", got, "こ")
+	}
+
+	if got := b.Bytes(); !bytes.Equal(got, []byte{0xE3}) {
+		t.Fatalf("buffer after second flush = %v, want %v", got, []byte{0xE3})
+	}
+
+	b.Write([]byte{0x82, 0x93})
+	if got := flushValidUTF8Prefix(&b); got != "ん" {
+		t.Fatalf("third flush = %q, want %q", got, "ん")
+	}
+
+	if b.Len() != 0 {
+		t.Fatalf("buffer not empty after third flush: %d", b.Len())
+	}
+}
+
+func TestFlushValidUTF8Prefix_ValidText(t *testing.T) {
+	var b bytes.Buffer
+	b.WriteString("hello 世界")
+
+	if got := flushValidUTF8Prefix(&b); got != "hello 世界" {
+		t.Fatalf("flush = %q, want %q", got, "hello 世界")
+	}
+
+	if b.Len() != 0 {
+		t.Fatalf("buffer not empty after flush: %d", b.Len())
+	}
+}