<xhash> should avoid including <xstring>

[StephanTLavavej]

Reported by: https://www.reddit.com/r/cpp/comments/wgb8xo/fun_times_with_msvc_functional/

This does two things:

• <xhash> includes <xstring> for basic_string, <cstring> for strlen, and <cwchar> for wcslen, because it's defining non-Standard stdext machinery for <hash_map> and <hash_set>. This throughput cost is paid by everyone who drags in <xhash>, including highly indirect routes like <functional>. There's a simple way to avoid this (i.e. without adding a new sub-header to be included by <hash_meow>). Because we hard-deprecated <hash_meow> with an error-and-escape-hatch, we can guard the stdext machinery with that same escape hatch. Result: the stdext machinery and <xstring> etc. dependencies disappear for the vast majority of users, and are still available unchanged when activating the escape hatch.
  • This throughput improvement is inherently a source-breaking change for anyone who assumed that <unordered_meow> would make most of <string> available. As usual, the fix is to simply include <string>. Let's be daring and try to see if we can do this without totally breaking the world.
• We were dragging stdext::hash_compare into std, contrary to our goal of keeping non-Standard identifiers out of std. The project that needed this, protobuf, was fixed almost 3.5 years ago. We should be able to eliminate this now.

[strega-nil-ms]

Thank you so much! This is great.

[StephanTLavavej]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.