<algorithm>: find and count could vectorize in more cases

[strega-nil-ms]

There are two cases which are currently not covered by _Vector_alg_in_find_is_safe and _Could_compare_equal_to_value_type/_Within_limits:

• is_pointer_v<_Elem> && is_null_pointer_v<_ValueType>
• is_pointer_v<_Elem> && is_pointer_v<_ValueType> && is_object_v<remove_pointer_t<_Elem>> && is_void_v<remove_pointer_t<_ValueType>>

[StephanTLavavej]

A couple of observations:

• This is symmetric, so in addition to searching for void* in a range of U*, the case of searching for U* in a range of void* could equally be optimized.
  • (There's no point in having symmetry for the nullptr_t case, of course 😹)
• Also, very much like the copy optimizations that we have, cv1 U* is guaranteed bit-identical to cv2 U*, so searching for one in a range of the other could be optimized. (That is, instead of insisting that the pointer types be absolutely identical, we could verify that they point to the same type after removing cv-qualifiers.)

There's definitely a complexity cost to adding additional metaprogramming and test coverage for such scenarios, so we need to draw the line at some point we consider reasonable - just some ideas for potential future work.