[MLInliner] bugfix for tensor type name: float -> float32

[yundiqian]

Summary

fix the bug that tensor type float is called float32 instead of float

[yundiqian]

mtrofin is going to fix in a different way.

llvm/include/llvm/Analysis/Utils/TFUtils.h

 
 /// List of supported types, as a triple:
 /// C++ type
-/// short name (for strings, for instance)
+/// TF tensor type string
 /// capitalized short name (for enums, for instance)
 #define TFUTILS_SUPPORTED_TYPES(M)                                             \
-  M(float, float, FLOAT)                                                       \
+  M(float, float32, FLOAT)                                                     \
   M(double, double, DOUBLE)                                                    \
   M(int8_t, int8, INT8)                                                        \
   M(uint8_t, uint8, UINT8)                                                     \