[Golang] Prevent OOM kills on corrupted data.

[dvyukov]

Most cases of corrupted data can be recovered.
However, OOM kills on very large allocations can't be recovered.
There is one case in object API that does make([]T) with size that comes
from the data. Add a sanity check before that allocation.

[dvyukov]

1. Is there a way to test this within the current test infrastructure?
2. Is there a better upper bound for array size?
3. If elements are guaranteed to be sequentially increasing in memory, then instead of:

	if fooLength > len(rcv._tab.Bytes)/4 {
		panic("bad array size")
	}
	t.Foo = make([]T, fooLength)
	for j := 0; j < fooLength; j++ {
		t.Foo[j] = rcv.Foo(j)
	}

It may be better to emit:

	if fooLength > 0 {
		_ = rcv.Foo(fooLength - 1)
	}
	t.Foo = make([]T, fooLength)
	for j := 0; j < fooLength; j++ {
		t.Foo[j] = rcv.Foo(j)
	}

[dvyukov]

It may be better to emit:

	if fooLength > 0 {
		_ = rcv.Foo(fooLength - 1)
	}
	t.Foo = make([]T, fooLength)
	for j := 0; j < fooLength; j++ {
		t.Foo[j] = rcv.Foo(j)
	}

Overall it looks better. But not sure about overflows here. Index functions don't have overflow checks and do calculations in UOffsetT, so even even on 64-bit arch multiplication in offset calculation can overflow.