xet-track llama-2.json and gpt2.json

Bench tokenizer configs for the pipeline-bench CI manifest (tk-encode/examples/bench_models.json): llama-2 = model-bounded BPE (no pre-tokenizer, byte_fallback), gpt2 = standalone ByteLevel pre-tokenizer.

Provenance (full unmodified tokenizer.json):
- `llama-2.json` from hf-internal-testing/llama-tokenizer @ main (tokenizer.json) — sha256 `8eea70c4866c4f1320ba096fc986ac82038a8374dbe135212ba7628835b4a6f1`
- `gpt2.json` from openai-community/gpt2 @ main (tokenizer.json) — sha256 `8414cab924d8b9b33013f0d221c5862f365ee9be39c5c2bfae8a5a9e970478a6`

.gitattributes

@@ -64,4 +64,6 @@ fixtures/**/*.txt filter=lfs diff=lfs merge=lfs -text
 bert-base-uncased.json filter=lfs diff=lfs merge=lfs -text
 deepseek-v4.json filter=lfs diff=lfs merge=lfs -text
 t5-base.json filter=lfs diff=lfs merge=lfs -text
+llama-2.json filter=lfs diff=lfs merge=lfs -text
+gpt2.json filter=lfs diff=lfs merge=lfs -text
 fixtures/models/** filter=lfs diff=lfs merge=lfs -text