FP in Java's BufferOverrun analysis with String.split(String, int) with non-positive limit

[friedbyalice]

Due to an incorrect modelling of the String.split(String, int) method, when the limit is 0 or -1, the analyzer assumes the resulting array will have a size [limit, 1]; according to JavaDoc, in this case the method will either

• behave like the String.split(String) overload, when the limit is 0
• in some cases return one more element than the overload, when the the string ends with a match, leading to an additional empty string not being split from the result; although I think treating 0 and -1 the same way, as if the 1 argument overload had been called, would be reasonable

For example

class Test {
  void method() {
    String s = "a,b";
    String[] split = s.split(",", -1);
    int[] _unused = new int[split.length];
  }
}

split.java:5: error: Alloc May Be Negative, Length: [-1, 1].
Reports INFERBO_ALLOC_MAY_BE_NEGATIVE on the integer array allocation.

I have a possible fix here, will open a PR after adding tests and some improvement (not sure why it would ever return [-1, 1], covering literals is not enough)