std.mem: Don't use vectorized path for freestanding targets in indexOfSentinel()

[alexrp]

This removes the invalid page size assumptions made in std.mem.indexOfSentinel() for freestanding targets. This is what originally prompted the commits that I'm reverting, which were added in #22488 and #23043. See the discussion: #22488 (review)

See also #23184.

[andrewrk]

What makes the page size assumptions invalid?

Also this is not a bug fix, this is breaking.

[alexrp]

What makes the page size assumptions invalid?

The key points from the previous discussion are:

• For freestanding, it's invalid to assume that paging is even enabled. But, even if it is, it may very well not be at the exact time that some standard library code such as std.mem.indexOfSentinel() is called.
• The standard library's notion of page size is not the same as the CPU architecture's. The former happens to be a more narrow subset of the latter because OSs are generally sane, but there is no iron law stating that this must be the case; you can create really weird illusions with page protection and code resumption tricks, for example.
  • The idea is to ultimately let the OS be the source of truth for what page sizes are possible, or in the absence of an OS, have the user tell us. This works out because, per the first point, we have to assume that paging is a meaningless concept on freestanding unless the user knows better.

Also this is not a bug fix, this is breaking.

Only 4f43918 needs to be picked into 0.14.1 to fix std.mem.indexOfSentinel() for freestanding. The other commits are just reverting the changes to std.heap.page_size_min/page_size_min_default because the case that prompted them is no longer relevant; these can just go in 0.15.0 since they're technically breaking. We're tagging 0.14.1 soon, so not including this anyway.