test_loadpaths_execute() was sometimes getting ERROR_ALREADY_EXISTS when
creating its temporary directoryi. That was probably because
GetTempFileNameW() only uses the low 16-bits of the 'random' LUID which
goes over a wider range, potentially leading to collisions in the lower
16-bits.
Avoid AllocateLocallyUniqueId() entirely and add a helper function to
create a guaranteed new directory.
Also try to use the current directory first as it is assumed to have
been configured to ward off anti-virus programs (specifically
Microsoft Defender) which is especially important when writing
executables to that location. Only switch to the system's temporary
directory if the current directory is not a writable location.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=54560

Today, RtlCreateActivationContext (CreateActCtxW) opens the source
manifest file via NtOpenFile without the FILE_SHARE_DELETE sharing mode.

This causes CreateActCtxW to fail if the source manifest file was
created with the FILE_DELETE_ON_CLOSE flag.  FILE_DELETE_ON_CLOSE is
often used for temporary files that should be automatically deleted
after use, even if the creator process crashes.

Fix this by specifying FILE_SHARE_DELETE for sharing mode when opening
the source manifest or module file.  This allows the source manifest or
module file to be marked as deleted while it is open.

Note that concurrent deletion is not an issue for the following reasons:

- The ability to read from an open file handle is unaffected by deletion
  of the corresponding file's name.

- RtlCreateActivationContext does not open the source manifest or module
  file by the given filename (lpSource) more than once.

When called with CERT_NAME_ATTR_TYPE and pvTypePara=NULL, Windows
did the first of email, CN, OU, or O while Wine just did email.

Fixed a regression from 9d537999.

And use it instead of next_thread_window_ptr.

Signed-off-by: Connor McAdams <cmcadams@codeweavers.com>

Signed-off-by: Connor McAdams <cmcadams@codeweavers.com>

Signed-off-by: Connor McAdams <cmcadams@codeweavers.com>

Signed-off-by: Connor McAdams <cmcadams@codeweavers.com>

The server is back online now and apparently its contents changed, causing
test failures.

advapi32:eventlog crashes the EventLog service in some Windows
versions so defend against that condition.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=53461

The ANSI one is broken in some UTF-8 locales which prevents most of the
tests from running. So use the Unicode API for most tests and just
confirm that it is consistent with the ANSI one instead of doing the
reverse.

Wine-Bug: https://bugs.winehq.org//show_bug.cgi?id=52895

Services sometimes start or stop between two EnumServicesStatus()
calls. Fortunately this is rare so the tests can just retry to get the
expected result.

Wine-Bug: https://bugs.winehq.org//show_bug.cgi?id=53460

It only has this one test.

The START_PENDING and STOP_PENDING services don't always have a
process id. Also trace unusual service states so we can identify which
ones behave in unexpected ways.

Wine-Bug: https://bugs.winehq.org//show_bug.cgi?id=53460

Fix the comments to not imply that the tests expect all services but
one to be returned.
Add a test of the needed buffer size to retrieve the remaining services
and document the limits of the test.

The pointers are used in loc_info input_name without copy.