CDRIVER-4781 Avoid intermediate uint32_t for socketTimeoutMS

[eramongodb]

Resolves CDRIVER-4781. Verified by this patch.

#1229 introduced range representability checks for conversion from int64_t -> int32_t timeout parameters in internal read/write utility functions, where due to historical reasons, int64_t timeout parameters for several read/write operations are forwarded to underlying int32_t timeout parameters (narrowing conversion). The new range checks are (correctly but unexpectedly) triggered by mongoc_cluster_t::sockettimeoutms when the URI option is set to -1, which has also been (implicitly) supported for historical reasons. The negative value was being implicitly converted to a large unsigned value before being passed as an int64_t, before being converted back to -1 when cast back to int32_t:

-1         (int32_t: MONGOC_URI_SOCKETTIMEOUTMS)
4294967295 (uint32_t: mongoc_cluster_t::sockettimeoutms)
4294967295 (int64_t: _mongoc_stream_writev_full())
  >int32_t range check failure< (the breaking change)
-1         (int32_t: mongoc_stream_writev(), impl-defined narrowing signed conversion)

This PR restores support for negative values for the socketTimeoutMS URI option for backward compatibility by replacing the intermediate uint32_t data members of mongoc_cluster_t with int32_t for better consistency with how the values are both assigned and used internally. This avoids the implicit conversion into a large unsigned value that triggers the range check before conversion back to an int32_t. This change also addresses 5 GCC and 5 Clang warnings for implicit sign conversion. Other instances of internal representation of timeout values (that aren't ignored, i.e. gridfs functions) do not seem to be affected: there are no other instances of intermediately-representing the timeout as an unsigned integer.

A regression test is added that ensures the range check is not triggered by a negative socketTimeoutMS URI value. This regression test + git bisect verified that the breaking change was triggered by abb0bb7 (#1229).

As a drive-by fix, this PR also fixes the incorrect format specifier for error messages concerning timeout_ms (PRId64 for int64_t).

Additionally, an attempt is made to document the unspecified/inconsistent behavior of non-positive timeout values for the URI option documentation and for functions that have a timeout parameter (i.e. mongoc_stream_*()). Scanning through how the timeout values are both represented and interpreted within libmongoc, it does not seem desirable to continue to support non-positive URI timeout values (even 0) despite historical (implicit) support.

[kevinAlbs]

LGTM with minor suggestions.