Update only Reference implementation and tests (#8126)

There were discovered a lot of issues in 'CompareWithRefs' tests when generated float inputs are not rounded to 'int'

ngraph/core/reference/src/runtime/reference/convert.cpp

@@ -60,7 +60,7 @@ void jit_convert_vec<float, int8_t>(jit::Generator& gen, const Xbyak::RegExp& sr
     auto p32vec_lo = gen.xmm2;
     auto p32vec_hi = gen.xmm3;
 
-    gen.vcvtps2dq(p32vec, gen.yword[src]);      // convert 8 floats to 8 ints
+    gen.vcvttps2dq(p32vec, gen.yword[src]);     // convert 8 floats to 8 ints
     gen.vpshufb(p32vec, p32vec, order);         // Shuffle the bytes according to the order
     gen.vextracti128(p32vec_hi, p32vec, 1);     // extract upper part of p32vec
     gen.vpor(p32vec_lo, p32vec_lo, p32vec_hi);  // p32vec_lo = p32vec_lo | p32vec_hi
@@ -80,7 +80,7 @@ void jit_convert_vec<float16, int8_t>(jit::Generator& gen, const Xbyak::RegExp&
     auto p32vec_hi = gen.xmm3;
 
     gen.vcvtph2ps(p32vec, gen.xword[src]);      // convert 8 fp16's to 8 floats
-    gen.vcvtps2dq(p32vec, p32vec);              // convert 8 floats to 8 ints
+    gen.vcvttps2dq(p32vec, p32vec);             // convert 8 floats to 8 ints
     gen.vpshufb(p32vec, p32vec, order);         // Shuffle the bytes according to the order
     gen.vextracti128(p32vec_hi, p32vec, 1);     // extract upper part of p32vec
     gen.vpor(p32vec_lo, p32vec_lo, p32vec_hi);  // p32vec_lo = p32vec_lo | p32vec_hi