Fixes github/codeql-c-team#553.

This PR does two things:

• The first commit removes a bunch of false positives by restricting the sinks to be signed integers in the case of potentially overflowing expressions. This removes a ton of results, and I think this is overall a good thing since (unlike signed-integer overflow) unsigned integer overflow is well-defined.

  Here's a list of 70 random projects that produce alerts on this query, and here is a difference query that shows how many results we remove with this PR.

  A note on Samate:
  On main this query raises 22798 TPs on Samate and after this PR it raises 18012 TPs.
  So we lose 4786 results. Out of those results, 1920 are related to CWE-680 (which we have a seperate query for), and the remaining 2866 results are CWE-190 related that look like:

  /* POTENTIAL FLAW: Use a random value */
  data = (unsigned int)RAND32();
  /* POTENTIAL FLAW: Adding 1 to data could cause an overflow */
  unsigned int result = data + 1;
  printUnsignedLine(result);

  which I'm not convinced is an issue that we want to highlight in real code.

• The second commit raises the query's precision to high, meaning it'll be part of the code-scanning suite and be displayed by default on LGTM.

Here's the results on our usual list of test projects: https://lgtm.com/query/7783254823209558200/.

I'm leaving this PR as a draft while I run the query on a larger set of LGTM projects.