Try to preserve metatable when exchanging data with the async env

[y5nw]

Add compact, short information about your PR for easier understanding:

This PR is similar to #14360 in terms of the goal (trying to preserve metatable information), but this one is implemented differently (and in a simpler way, as certain cases in minetest.serialize are irrelevant here). The two PRs are otherwise not related to each other.

• Goal of the PR
  Preserve metatable information when passing data to/from the async environment
• How does the PR work?
  It keeps a mapping between certain metatables and a name, the latter of which is kept with the data when transferred to a different environment and then used to set the metatable on the reconstructed data.
  Note that, unlike Try to preserve metatable information in minetest.serialize #14360, this PR only handles tables. It does not handle userdata objects as these are handled by C++ code directly.
• Does it resolve any reported issue?
  Fixes Vector metatable doesn't carry over to async environment #13644.
• Does this relate to a goal in the roadmap?
  2. Bugfix; the issue linked above is labeled as a bug.

To do

This PR is Ready for Review.

• Metatables are preserved.
• Add documentation.

How to test

The base test case for vectors should be covered by the unittests in the Devtest.

To test the modding API, use the following mod:

-- init.lua
local MP = minetest.get_modpath(minetest.get_current_modname())

dofile(MP .. "/pair.lua")
minetest.register_async_dofile(MP .. "/pair.lua")

local list = pair(1, pair(2, nil))
minetest.handle_async(function(x)
	return x
end, function(lst)
	print(list, lst, list == lst)
end, list)

-- pair.lua
local mt = {
	__eq = function(x, y)
		return x[1] == y[1] and x[2] == y[2]
	end,
	__tostring = function(x)
		return ("(%s . %s)"):format(x[1], x[2])
	end,
}

function pair(x, y)
	return setmetatable({x, y}, mt)
end

minetest.register_async_metatable("async_test:pair", mt)

The above mod should print a line to the console:

(1 . (2 . nil)) (1 . (2 . nil)) true

[y5nw]

Rebased.

[sfan5]

tested and works
(if the CI fails now that would be embarassing)

[SmallJoker]

Discussed in the meeting. https://irc.minetest.net/minetest-dev/2024-03-03#i_6156924

Contains more ideas for possible optimizations (not for this PR).

[appgurueu]

Thanks for the PR.

The code looks correct. Unit tests pass. I took the liberty to extend the async metatable registration test with a table with a custom, registered metatable.

My comments are nitpicks; do what you want with them.

I rebased because otherwise it wouldn't compile.

[sfan5]

With #14451 it will become relevant to extend this mechanism to the mapgen env too.
I propose a rename to minetest.register_portable_metatable.

[y5nw]

With #14451 it will become relevant to extend this mechanism to the mapgen env too.
I propose a rename to minetest.register_portable_metatable.

I renamed this to register_metatable in #14360, but register_portable_metatable is IMO better.