Correctly compute field path for the heap snapshot, taking struct inlining into account

[NHDaly]

The existing debug code for computing field name from a slot pointer is not working: gc_slot_to_fieldidx(parent, slot).

I think this is because that code was written before the optimization that was added in julia 1.3 or 1.4 to allow inlining immutable structs with heap references.

This PR creates a new function, _fieldpath_for_slot(from, slot) which returns a vector of type, field pairs, that give the full path to a slot's field from a given object.

For example, for the following struct:

struct B
    x::Int
    y::Any
end
mutable struct A
    a::Int
    b::B
end

ga = A(1, B(2, []))

The only pointer field in ga is: b.y :: Vector{Any}, which is returned as [(B, :y), (A, :b)] (in reverse order because of silly implementation reasons, since C++ vectors aren't efficient for push_front).

[NHDaly]

@vilterp i think this LGTM. Do you want to have a look at it?

[vilterp]

I honestly don't mind the combined path 🤔 🤷‍♂️

[NHDaly]

yeah, i'm not sure what to think about it either. After we talked, I tried adding the (b::B).y type syntax, but it's a bit difficult / awkward, so i didn't 😅. String processing in C++ is TOO HARD