Add options to use scons caching for faster iteration.

[PapyChacal]

While working on godot-cpp I got tired of recompiling every object file when changing only some files' generation logic.

Thus this Pull Request: I added options to the SConstruct to leverage scons' caching capabilities. If activated, it stores every build output in the cache, and fetch it if the dependencies have the exact same content. With the option enable, recompiling after binding regeneration will only recompile objects where the generated sources have actually changed, and fetch all others from the cache.

This required to sort the sets used in binding_generator.py: otherwise the includes and forward-dcelaration are done in pseudo-random order, leading to different content despite being equivalent.

I didn't notice any impact on the generation time, but if this is a worry I can sort them only when caching is enabled.

[PapyChacal]

I adding a dependency on the source generation too: now any modification to binding_generator.py triggers the source generation, as well as modifying extension_api.json or the header, meaning we don't have to use generate_bindings=yes so much.

I also enforced source generation when using generate_bindings=yes, cache or no cache.

[Faless]

Looks really good 🏆 , see review comments, there's also an extra deleted line before add_sources which makes CI fail.

[Faless]

One more style change required by static checks.

Also needs squashing, see https://docs.godotengine.org/en/latest/community/contributing/pr_workflow.html#the-interactive-rebase

[Faless]

One last change required by static checks (python formatting):

Suggested change

	[json_api_file, os.path.join(env["headers_dir"], "godot", "gdnative_interface.h"), "binding_generator.py"]
	[json_api_file, os.path.join(env["headers_dir"], "godot", "gdnative_interface.h"), "binding_generator.py"],

[PapyChacal]

Just out of curiosity if you have time, I find this comma unsettling. What is the rationale for requiring it? (I added it anyway in the squash)

[Faless]

It's the style enforced by black https://pypi.org/project/black/ .
I think it defaults to that style for functions where the ending parenthesis is on a new line.
I don't particularly like that style either, but it's totally valid python.

[PapyChacal]

Last formatting and squash done, thanks!

[Faless]

Thanks!