Allocation sites sized by definite-length array types should be recorded another way

[stephenrkell]

This problem was witnessed in the libuv codebase (thanks @mbyzhang).

If we size an allocation like this (in libuv's test suite, specifically test-fs.c):

static char test_buf[] = "test-buffer\n";
...
buffer = malloc(sizeof(test_buf) * iovcount);

then the allocation type will be recorded as "array of 13 chars". I think it should just be "char" and we are not stripping arrays in dumpallocs.ml.

This is going wrong in allocsites because the tool is searching for the exact DWARF type signed char [13] which is not directly embodied... specifically it is not in the types_by_codeless_name relation that we built. I am not sure why it is not, however. Presumably it should be there? If the debug info for test_buf is intact, then presumably it is in the info section. So I'd like to understand this one a little better before jumping to a fix.

[stephenrkell]

In #182 we fixed the name-mangling issue, but the question of what to do about definite-length array types remains. I think the allocation site record should have an additional field, sized_in_multiples_of that captures any stripped array length, while the main type field will just record "array of" (as per #34).