[CDRIVER-5859] Enable sign-compare warnings globally, and fix them all

[vector-of-bool]

Refer: CDRIVER-5859 This changeset adds sign-compare warnings at the top-level, and fixes every occurrence. Most occurrences just swap for the new mlib_cmp macro, but a few more changes have been made to clear up sign-handling.

Summary

• Lots of use of mlib_cmp, from PR CDRIVER-5859 Integer comparison macro #1845
• c763a73: A statement-macro that tests whether a given (compound) statement raises SIGABRT. This is a no-op on Windows. This uses fork without exec, so it isn't super fast, usually incurring a few milliseconds for each use.
• e830cf5: A macro for looping over a half-open range of integers.
• 93efe4d: A macro for looping over arrays.
• d01e7f6: A new header: mlib/intencode.h, which adds functions dedicated to decoding/encoding integer values into raw memory.
• b689d49: This commit actually fixes all instances of sign-comparison warnings in the codebase.

[eramongodb]

Very nice changes so far.

Some additional suggestions:

• Also replace or remove use of BSON_MIN and BSON_MAX from <bson/bson-macros.h> to address the following warning(s) due to the internal comparison expression:

src/libbson/src/bson/bson-macros.h:121:30: warning: comparison of integer expressions of different signedness: ‘unsigned int’ and ‘int’ [-Wsign-compare]
  121 | #define BSON_MIN(a, b) (((a) < (b)) ? (a) : (b))
      |                              ^
src/libbson/src/bson/bson-decimal128.c:303:38: note: in expansion of macro ‘BSON_MIN’
  303 |          const unsigned n_to_write = BSON_MIN (n_trailing_digits, available_bytes);
      |                                      ^~~~~~~~
src/libbson/src/bson/bson-macros.h:121:45: warning: operand of ‘?:’ changes signedness from ‘int’ to ‘unsigned int’ due to unsignedness of other operand [-Wsign-compare]
  121 | #define BSON_MIN(a, b) (((a) < (b)) ? (a) : (b))
      |                                             ^~~
src/libbson/src/bson/bson-decimal128.c:303:38: note: in expansion of macro ‘BSON_MIN’
  303 |          const unsigned n_to_write = BSON_MIN (n_trailing_digits, available_bytes);
      |                                      ^~~~~~~~

• Also update ASSERT_CMPINT_HELPER in TestSuite.h to use mlib_cmp for its internal equality comparison. Note the non-integer ASSERT_CMP* macros currently implemented in terms of ASSERT_CMPINT_HELPER (for double and void*) will need to be special-cased to avoid using incorrect use with the new integer comparison utilities.

[vector-of-bool]

@eramongodb

• I considered how to modify or replace uses of BSON_MIN/BSON_MAX, but it's difficult in that one cannot statically determine whether the caller intends to have a signed/unsigned result in the case that the signs mismatch. For now, I just let -Wsign-compare find the issues and then update the call sites so that the signs agree with what the caller actually intends to happen.
• I can update ASSERT_CMP* functions to use this compare, but I could also replace all uses of ASSERT_CMP<int-type> with mlib_check, which will do the correct assertion with a diagnostic. This will touch hundreds of lines, but significantly simplifies integer assertions. Do you (or anyone else) have any opinions on whether to do this swap?

[kevinAlbs]

The consistent use of helpers to read/write integers is much appreciated.

[eramongodb]

Do you (or anyone else) have any opinions on whether to do this swap?

I am fine with deferring those changes to a followup PR.

[kevinAlbs]

Suggested change

	*(_list).current++ = (_bson_append_bytes_arg) { \
	*(_list).current++ = (_bson_append_bytes_arg){ \

To fix clang-format task. Can be run locally with:

./tools/poetry.sh run .evergreen/scripts/clang-format-all.sh

[vector-of-bool]

Uggh. My local clang-format is updated past the version installed by Poetry, and it now adds a space after compound-initializers every time I save a file. I need to figure out what .clang-format setting does this, because it is a chore to remember not to commit them.

[vector-of-bool]

It looks like clang-format 19 now treats compound initializers as C-style casts, so our SpaceAfterCStyleCast option now inserts a space. Earlier Clang versions don't seem to have a way to control the whitespace here, and always deleted it.