[SYCL][Joint matrix] clarify the range of the prefetch templated argu…

…ments (#13796)

This is a clarification to match all the matrix spec that `Rows` and
`Cols` are in the range of the logical dimensions, independently from
the memory layout.
For instance, in the `ext_intel_packed` (or VNNI) layout case, prefetch
will look like:
`joint_matrix_prefetch<TK, TN>(sg, B + sg_starty / SG_SZ * TN *
vnniFactor,
N * vnniFactor, layout::ext_intel_packed,
syclex::properties{syclex::prefetch_hint_L1});
`
instead of 
`joint_matrix_prefetch<TK / vnniFactor, TN * vnniFactor>(
sg, B + sg_starty / SG_SZ * TN * vnniFactor, N * vnniFactor,
layout::ext_intel_packed,
syclex::properties{syclex::prefetch_hint_L1});
`

sycl/doc/extensions/experimental/sycl_ext_matrix/sycl_ext_oneapi_matrix.asciidoc

@@ -467,6 +467,10 @@ q.parallel_for(nd_range<2>(G, L), [=](nd_item<2> it) {
   }
 });
 ```
+The `Rows` and `Cols` template parameters represent the logical shape
+of the matrix, which is not necessarily the way the matrix is layed
+out in memory. Thus, these template parameters have the same meaning
+in `joint_matrix_prefetch` as they do for `joint_matrix_load`.
 
 === Support for Machine Learning Types
 Some devices support special matrix element types that are commonly