Address feedback comment on using familar MLIR syntax for eBNF grammar

docs/spec.md

@@ -146,20 +146,21 @@ IntegerType ::= 'si4' | 'si8' | 'si16' | 'si32' | 'si64'
               | 'ui4' | 'ui8' | 'ui16' | 'ui32' | 'ui64'
 FloatType   ::= 'f8E4M3FN' | 'f8E5M2' | 'bf16' | 'f16' | 'f32' | 'f64'
 ComplexType ::= 'complex' '<' ('f32' | 'f64') '>'
-QuantizedType ::= '!quant.uniform' '<' QuantizationStorageType
+
+QuantizedType ::= '!quant.uniform' '<' QuantizationStorageTypeSpec
                   ':' QuantizationExpressedType
-                  ':' QuantizationStorageMin
-                  ':' QuantizationStorageMax
-                  ':' QuantizationScales
-                  ':' QuantizationZeroPoints
-                  {':' QuantizationDimension} '>'
+                  [':' QuantizationDimension]
+                  ',' QuantizationScalesAndZeroPoints '>'
+QuantizationStorageTypeSpec ::= QuantizationStorageType ['<' QuantizationStorageMin ':' QuantizationStorageMax '>']
 QuantizationStorageType ::= IntegerType
-QuantizationExpressedType ::= FloatType
 QuantizationStorageMin ::= IntegerConstant
 QuantizationStorageMax ::= IntegerConstant
-QuantizationScales ::= '[' FloatConstant {',' FloatConstant} ']'
-QuantizationZeroPoints ::= '[' IntegerConstant {',' IntegerConstant} ']'
+QuantizationExpressedType ::= FloatType
 QuantizationDimension ::= IntegerConstant
+QuantizationScalesAndZeroPoints ::= (QuantizationScaleAndZero | '{' QuantizationScaleAndZero {',' QuantizationScaleAndZero} '}')
+QuantizationScaleAndZero ::= QuantizationScale ':' QuantizationZeroPoint
+QuantizationScale ::=  FloatConstant
+QuantizationZeroPoint ::=  IntegerConstant
 ```
 
 **Element types** represent elements of tensor types. Unlike in many programming
@@ -215,21 +216,21 @@ Quantized types satisfy the following constraints:
     * `storage_min = -2^(d-1)+1`, `storage_max = 2^(d-1)-1`.
   * If `is_unsigned(storage_type)`:
     * `storage_min = 0`, `storage_max = 2^d - 1`.
-* (C5) For all `i`, `type(scales[i]) = expressed_type`.
-* (C6) For all `i`, `scales[i] > 0`.
-* (C7) For all `i`, `type(zero_points[i]) = i64`.
-* (C8) `size(scales) = size(zero_points)`.
-* (C9) If `quantization_dimension` is empty, then `size(scales) = 1`.
-* (C10) If `quantization_dimension` is not empty, then
+* (C5) `type(scale) = expressed_type`.
+* (C6) `scale > 0`.
+* (C7) `type(zero_point) = i64`.
+* (C8) If `quantization_dimension` is empty, then
+  `size(scales_and_zero_points) = 1`.
+* (C9) If `quantization_dimension` is not empty, then
   `0 <= quantization_dimension`.
 
 Furthermore, tensors of quantized types satisfy the following constraints:
 
 * For per-tensor quantization:
   * No additional constraints.
 * For per-axis quantization:
-  * (C11) `quantization_dimension < size(shape)`.
-  * (C12) `size(scales) = shape[quantization_dimension]`.
+  * (C10) `quantization_dimension < size(shape)`.
+  * (C11) `size(scales_and_zero_points) = shape[quantization_dimension]`.
 
 ```ebnf
 FunctionType ::= '(' [ValueType {',' ValueType}] ')' '->' '(' [ValueType {',' ValueType}] ')'