[oneDPL] Expand and clarify differences with serial range algorithms (#610)

Author: akukanov

source/elements/oneDPL/source/parallel_api/parallel_range_api.rst

@@ -23,7 +23,15 @@ The following differences to the standard C++ range algorithms apply:
 - The execution policy parameter is added.
 - Output data sequences are defined as ranges, not iterators.
 - Both input and output ranges must support random access.
-- For a given algorithm, at least one of the input ranges as well as the output range must be bounded.
+- As a rule, both input and output ranges must be sized.
+
+  - Exceptions are binary ``transform``, ``equal``, and ``mismatch``, where at least one of the input ranges
+    must be sized, and if a range is not sized it is supposed to be infinite.
+    [*Note*: An example of an infinite range is ``std::views::repeat`` with no bound. -- *end note*]
+
+- For algorithms with bounded output ranges, processing may not need to go over all the input data.
+  In that case, the returned value usually contains iterators pointing to the positions past the last elements
+  processed according to the algorithm semantics.
 - ``for_each`` does not return its function object.
 
 Except for these differences, the signatures of parallel range algorithms correspond to the working draft