Don't throw resize lock exception in destructor when removing nodemeta inventory

[Desour]

If you remove an inventory in meta in one of its inv callbacks, minetest currently crashes due to a resizelock exception in a destructor.
This PR triggers the exception before calling the destructor, so there will only be a lua error.
(The nodemeta will be left half-deleted. Only has the inventory will remain remain, not meta fields or privateness info.)

Fixes #13785 (comment).

To do

This PR is a Ready for Review.

How to test

minetest.register_node("test_inv_callbacks:node", {
	description = "inv callback node",
	tiles = {"default_cobble.png^heart.png"},
	groups = {choppy = 3, oddly_breakable_by_hand = 3},

	on_construct = function(pos)
		minetest.log("on_construct")
		local meta = minetest.get_meta(pos)
		local inv = meta:get_inventory()
		inv:set_size("mylist", 5)
		meta:set_string("formspec", "size[8,10]"
			.."list[current_player;main;0,3;8,4;]"
			.."list[current_name;mylist;0,0;10,1;]")
	end,
	on_metadata_inventory_move = function(pos, from_list, from_index,
			to_list, to_index, count, player)
		minetest.log("on_metadata_inventory_move")
		core.remove_node(pos)
	end,
})

[grorp]

Since the on_* inventory callbacks are called after the items are moved, not before or while they are moved, this shouldn't even result in a Lua error, should it?

minetest/doc/lua_api.md

Lines 9206 to 9211 in 6fdc7e0

	on_metadata_inventory_move = function(pos, from_list, from_index, to_list, to_index, count, player),
	on_metadata_inventory_put = function(pos, listname, index, stack, player),
	on_metadata_inventory_take = function(pos, listname, index, stack, player),
	-- Called after the actual action has happened, according to what was
	-- allowed.
	-- No return value.

Is it possible to already have the resize lock released at this point?

[Desour]

Is it possible to already have the resize lock released at this point?

In some cases, probably.
Currently the list_from is kept lock for the case where the move/put/take is a swap, to get the item for the 2nd callback.

Anyway, the crash also happens if the node is removed in an allow_ callback.

[grorp]

Works, but doesn't fully resolve #13785 yet.

[SmallJoker]

@grorp Your theory does not hold true in the case where ItemStacks are limited in count due to allow_* callbacks. When shift-clicking (i.e. moves according to listring[]), this would split the stack, resulting in subsequent callbacks.