CDRIVER-5634 account for embedded NULL when copying salt

[kevinAlbs]

Followup to #1684. Fixes test failures on Windows with SCRAM auth. Example failure:

Begin /scram/cache_invalidation, seed 1723487056
    { "status": "fail", "test_file": "/scram/cache_invalidation", "seed": "1723487056", "start": 965.718000, "end": 974.359000, "elapsed": 8.641000  },
FAIL:C:\data\mci\26ca9ede4b9bb4808634a41a0bd207c0\mongoc\src\libmongoc\tests\test-mongoc-scram.c:457  _try_auth_from_uri()
  res
  Authentication failed.

Verified with this patch build.

Running /scram/cache_invalidation (which auths many times) on a spawn host showed salt may contain NULL bytes. This resulted in truncation when salt is copied with bson_strndup. This PR changes the copy to use bson_malloc + memcpy.

I expect embedded NULLs are not permitted in passwords, but copying the password was updated similarly for consistency.

[eramongodb]

Suggested change

	char *password_copy = bson_malloc (sizeof (char) * password_len);
	memcpy (password_copy, password, password_len);
	uint8_t *salt_copy = bson_malloc (sizeof (uint8_t) * salt_len);
	memcpy (salt_copy, salt, salt_len);
	unsigned char *password_copy = bson_malloc (password_len);
	memcpy (password_copy, password, password_len);
	unsigned char *salt_copy = bson_malloc (salt_len);
	memcpy (salt_copy, salt, salt_len);

Suggest taking the opportunity to "cast" to unsigned char during the memcpy operations and remove the (unsigned char*) casts below.

sizeof (char) is guaranteed to be 1. We do not support (or I believe plan to support) any target where sizeof (uint8_t) != 1.