Refactor variant_call.cpp to make better use of templates, eliminating one layer of abstraction

[Ivorforce]

This PR simplifies some functions in variant_call.cpp. In particular:

• #define METHOD_CLASS and others are replaced with templated register_method and others.
• vc_get_argument_type and vc_get_argument_count are replaced with vc_get_argument_types.

The goal of this PR is to reduce complexity in the file by:

• Turning more #define wrapped code into 'real' code, making it easier to edit and understand.
• Removing a layer of abstraction between calling a register macro and registering the function.

My ulterior motive is to eventually propose exposing this API to GDExtension. But this PR should be defendable without this motive :)

Benchmarks

I do not expect either an improvement or regression, but it should be tested to be sure about it.
I benchmarked editor launch as a quick and dirty entry test for this:

 ❯ hyperfine -m25 -iw1 "bin/godot.macos.editor.arm64.master --path /Users/lukas/dev/godot/empty-game --editor --quit" "bin/godot.macos.editor.arm64 --path /Users/lukas/dev/godot/empty-game --editor --quit"
Benchmark 1: bin/godot.macos.editor.arm64.master --path /Users/lukas/dev/godot/empty-game --editor --quit
  Time (mean ± σ):      4.223 s ±  0.110 s    [User: 2.770 s, System: 0.363 s]
  Range (min … max):    3.990 s …  4.359 s    25 runs

Benchmark 2: bin/godot.macos.editor.arm64 --path /Users/lukas/dev/godot/empty-game --editor --quit
  Time (mean ± σ):      4.192 s ±  0.106 s    [User: 2.778 s, System: 0.363 s]
  Range (min … max):    3.978 s …  4.342 s    25 runs

Summary
  bin/godot.macos.editor.arm64 --path /Users/lukas/dev/godot/empty-game --editor --quit ran
    1.01 ± 0.04 times faster than bin/godot.macos.editor.arm64.master --path /Users/lukas/dev/godot/empty-game --editor --quit

I'm taking this as no difference (change within expected fluctuations).

I also benchmarked the unit tests:

 ❯ hyperfine -m5 -iw1 "bin/godot.macos.editor.arm64.master --test" "bin/godot.macos.editor.arm64 --test"
Benchmark 1: bin/godot.macos.editor.arm64.master --test
  Time (mean ± σ):     13.234 s ±  0.113 s    [User: 6.353 s, System: 0.843 s]
  Range (min … max):   13.051 s … 13.326 s    5 runs

Benchmark 2: bin/godot.macos.editor.arm64 --test
  Time (mean ± σ):     13.086 s ±  0.056 s    [User: 6.333 s, System: 0.817 s]
  Range (min … max):   13.029 s … 13.164 s    5 runs

Summary
  bin/godot.macos.editor.arm64 --test ran
    1.01 ± 0.01 times faster than bin/godot.macos.editor.arm64.master --test

The difference is similar as before, so probably within expected fluctuations.

A better test would be benchmarking actual calls (call, validated_call and ptrcall). I might do this, but I'd also appreciate if someone else stepped up!

Size Decrease

As a side effect, the size of the final binary shrinks by about 0.8mb (0.5%).
I checked with bloaty about this, and the decrease is indeed caused by the templates taking up less space than the structs we had before:

 ❯ diff <(bloaty bin/godot.macos.editor.arm64.master -d symbols -- -n 500) <(bloaty bin/godot.macos.editor.arm64 -d symbols -- -n 500)
3,4c3,4
<   41.3%  68.4Mi  41.0%  68.8Mi    [116355 Others]
<   10.2%  16.9Mi  10.1%  16.9Mi    [__LINKEDIT]
---
>   41.0%  67.7Mi  40.7%  68.0Mi    [112454 Others]
>   10.2%  16.8Mi  10.1%  16.8Mi    [__LINKEDIT]
[...]
<    0.2%   395Ki   0.2%   395Ki    _register_variant_builtin_methods_array()
58d57
<    0.2%   311Ki   0.2%   311Ki    _register_variant_builtin_methods_math()
74d72
<    0.1%   251Ki   0.1%   251Ki    _register_variant_builtin_methods_string()
89d86
<    0.1%   201Ki   0.1%   201Ki    _register_variant_builtin_methods_misc()
104a102
>    0.1%   178Ki   0.1%   178Ki    _register_variant_builtin_methods_array()
127a126
>    0.1%   159Ki   0.1%   159Ki    register_convert_method<>()::{lambda()#1}::__invoke()
145a145
>    0.1%   142Ki   0.1%   142Ki    _register_variant_builtin_methods_string()
148a149
>    0.1%   134Ki   0.1%   134Ki    _register_variant_builtin_methods_math()
150a152
[...]
<  100.0%   165Mi 100.0%   167Mi    TOTAL
---
>  100.0%   164Mi 100.0%   166Mi    TOTAL

[Calinou]

Tested locally, it works as expected. Code looks good to me.