[DRAFT] [inferbo] give global and static arrays their declared sizes

[VladimirMakaev]

Inferbo knew a global array's size only from its initializer, so it missed overruns of
uninitialized global arrays (#949), of static local arrays and of array fields of global structs.
A store g[i] = v also overwrote the whole array value of g.

int arr[50];

void write_all() {
  for (int i = 0; i <= 50; i++) {
    arr[i] = 1; // Infer missed: buffer overrun
  }
}

The clang frontend records the declared types of the globals each procedure references.
Inferbo gives these arrays symbolic contents of that size, stores update their elements, a row
g[i] used as int* points to its first element, and only const arrays take their contents
from their initializer.

Limitations:

• array fields of global structs passed to calls or strcpy need [DRAFT] [inferbo] bounds-check array fields of local structs and objects #2185 too (together: Array size in struct or global scope not captured? #1444);
• elements are weak (if (g[1] > 3) bounds no later read) and conditions on them are reported
  in each caller, as for other globals;
• pointers to rows (int (*q)[4] = &g[1]; (*q)[3]) keep Inferbo's imprecision for local arrays.

Test plan

New tests in c/bufferoverrun/global.c, cpp/bufferoverrun/global.cpp,
cpp/bufferoverrun/std_array.cpp and c/performance/loops.c; issue949_bad_FN becomes
issue949_bad. The addresses of two mutable global arrays are no longer known to differ, which
adds 1 to the cost of loop_use_global_vars (C and ObjC). The C/C++ codetoanalyze tests pass.