fix(xnnpack): re-export int8 with the get_model_schema flow

The published int8 .pte predated the schema flow: it carried no get_model_schema method and a [1,128,vocab] decode contract, so react-native-executorch rejected it at load ('Cannot match symbolic dynamic with concrete constant'). Re-exported from export-scripts main; encode now declares a 201..480000 range and decode [1,1,vocab], matching the fp32 variant. Verified by transcription.

xnnpack/whisper_tiny_en_xnnpack_int8.pte

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:445e0ab2bbeee72da73c7d8301855e9f59d68893ae5ded98289516914f015652
-size 175868672
+oid sha256:d006da76b6c3b7e5e82e4a3fe33551c29ca2ff5c73e07ffe264f20d6b8bf2bdb
+size 175347072