Imprecise cast overflows in sv-benchmarks

[sim642]

Regarding the question whether casts are overflows in SV-COMP, I ran Goblint both ways and inspected the verdict differences. There are a bunch of tasks where it seems that we are just rather imprecise if casts were to be counted as overflows:

• Juliet_Test/CWE190_Integer_Overflow__int_connect_socket_*_good (152 tasks): implicit cast ssize_t (long int) -> int on LP64, Goblint imprecise: recv doesn't return bigger than len argument
• Juliet_Test/CWE190_Integer_Overflow__int_listen_socket_*_good (152 tasks): implicit cast ssize_t (long int) -> int on LP64, Goblint imprecise: recv doesn't return bigger than len argument
• Juliet_Test/CWE190_Integer_Underflow__int_{connect,listen}_*_good (304 tasks): likely analogous to the Overflow cases, didn't check
• bitvector/gcd_* (3 tasks): implicit cast signed char -> int and int -> signed char on ILP32
• bitvector/s3_srvr_2a*.BV.c.cil (2 tasks): explicit cast int -> unsigned int and unsigned int -> int on ILP32
• loop-invariants/even
• array-crafted/bAnd* (5 tasks)
• array-crafted/bor* (5 tasks)
• array-crafted/xor* (5 tasks)
• array-memsafety/cstrcspn-alloca-1
• array-memsafety/cstrlen-alloca-2
• array-memsafety/cstrspn-alloca-1
• aws-c-common/aws_array_list_swap_harness

If the SV-COMP community decides that casts are overflows, we might want to fix these imprecisions.