https://github.com/llvm/llvm-project/issues/159474

Another try of trying to land
https://github.com/llvm/llvm-project/pull/166382
- Fix some leftover tests checking for specific  errnos
- Guard errno checking tests to not run on the GPU

@michaelrj-google

https://github.com/llvm/llvm-project/issues/159474

Resubmitting https://github.com/llvm/llvm-project/pull/162876 with fixes
as it broke some buildbots:
 - Fix comparisons of integer expressions of different signedness
- Not check for specific errnos in tests, as they might not be available
on all platforms

[#159474](https://github.com/llvm/llvm-project/issues/159474)

- All printf variants set errno and consistently return -1 on error,
instead of returning various predefined error codes
- Return value overflow handling is added

Tests can be at top-level or inside an anonymous namespace,
doesn't matter.  But putting their helper code inside anonymous
namespaces both makes the code compatible with compiling using
-Wmissing-declarations and might let the compiler optimize the
test good a bit better.

Fixes a couple of bugs found when building. The PR to enable the headers
can be found here: #144114.

- math.yaml: float128 guard
- wchar.yaml: __restrict keyword order

Summary:
Currently we dispatch the writing mode off of a runtime enum passed in
by the constructor. This causes very unfortunate codegen for the GPU
targets where we get worst-case codegen because of the unused function
pointer for `sprintf`. Instead, this patch moves all of this to a
template so it can be masked out. This results in no dynamic stack and
uses 60 VGPRs instead of 117. It also compiles about 5x as fast.

Relates to
https://github.com/llvm/llvm-project/issues/119281#issuecomment-2699470459

This reverts commit 1e6e845d49a336e9da7ca6c576ec45c0b419b5f6 because it
changed the 1st parameter of adjust() to be unsigned, but libc itself
calls adjust() with a negative argument in align_backward() in
op_generic.h.