Make inst2dict and dict2inst work recursively

[Xrayez]

Closes #6533.

This makes inst2dict GDScript method to turn inner instances present as properties to be converted to a dictionary recursively, including instances with nested arrays and dictionaries with key/value pair.

Added a deep paramater to inst2dict so that the default behavior remains the same. Recursive conversion is disabled by default because instances may have circular references as described in #31229 (comment), hence must be used with caution on case-to-case basis. Updated documentation accordingly.

Features

• Convert properties with instances (inner classes) recursively using inst2dict.
• Convert nested arrays with classes recursively using inst2dict.
• Convert nested dictionaries with classes recursively.
• Convert all back using dict2inst (all above).
• Add recursive/deep option for inst2dict method.

Test project

Extensive test project demonstrating all of the added features, and showing how having circular references fail, which is handled in debug builds by checking that only unique objects are converted.

inst2dict-recursive.zip

[Xrayez]

While testing serialization features with:

var json = to_json(inst2dict(inst, true)) # recursive
var inst = dict2inst(parse_json(json))

I stumbled upon serialization "bug" pertaining to JSON parsing dictionary keys as described in #33161. If this feature gets merged then it certainly must be fixed.

[Xrayez]

An alternative fix is to not turn dictionary keys (which can be object instances) into "dictionary instances" as seen in:

for (List<Variant>::Element *E = keys.front(); E; E = E->next()) {
	// Both key and value can be objects
	Variant vi = _inst_to_dict_or_var(dict[E->get()], r_error, p_deep);
	// Variant ki = _inst_to_dict_or_var(E->get(), r_error, p_deep); // do not support it?
	Variant ki = E->get()
	ret_dict[ki] = vi;
}

But these can lead to unexpected serialization issues. Or just add explicit error check and call it a feature.

[Xrayez]

Note: this was initially developed for/against 3.2 and was well tested there at the time of writing (see test project).

@vnen given there was a "duplicate" PR fixing the same issue (#38377), I wonder what would be a good time to make this PR work against current 4.0 master branch, and whether this could be backported to 3.2 once available in master (if the feature is desired at all).

[Xrayez]

Rebased, but the test project fails now in current master f8c066e, will need to investigate further... This worked fine in 3.2, not sure what could change since that time.

This is probably related to broken dictionaries as described in the similar the PR #38377 by @ThakeeNathees, but I'm not sure. Somehow, all the properties get converted to null on dict2inst, while inst2dict works fine 😕.

Not sure whether I'm motivated enough to make this work, and it would likely be safer to redo this from anew against master branch anyways to make sure that the rebase didn't lose the previous smaller fixes as seen in the tracker: #33348 (too many delete/add code due to moving implementation to methods), so I better close this.

In fact, I don't mind if @ThakeeNathees could get some insights from this PR, take both test projects and see how this can be improved, the idea is pre-approved by @vnen (IRC log):

vnen> Xrayez: I think the idea is fine, though it should be reviewed. On a quick glance it looks fine. If you fix the conflicts we can review and merge

If there's still no work done by any of the potential contributor, I might return to this when I'll actually need this feature myself.

I've saved the original branch which was originally targeted for 3.2: https://github.com/Xrayez/godot/commit/9ea46abcc42b000776f5c5a5d1487fae4d10db06.