v0.2.0-beta2-26.3+neoforge

As this is still a beta version, this version can contain bugs. Feel free to report ANY misbehaviours and errors!

• the GPU part transform costs the GPU less without a shader pack: on Intel's Linux driver (Mesa) its compute pass split every frame into three GPU submissions that wait for each other, a fixed cost however few models it draws. Where the GPU allows it, the pass now runs as a vertex shader inside the frame's normal submission: on an Intel Arc 140V 0.03 instead of 0.24 ms of GPU time per frame in a scene with only a few animated models, and frames 2–7 % faster than with 0.2.0-beta1's compute pass at 1920×1080. With a shader pack it stays a compute pass as in 0.2.0-beta1 (there both are equally fast at 1920×1080, and the compute pass is faster in a smaller window). Both write exactly the same vertices; F3 mobium:modules shows which one runs, and -Dmobium.parts.stage=compute or vertex in the Java arguments picks one
• /mobium stages [seconds] [runs] [scene] compares the two ways of running the GPU part transform in one session; benchmark setup names get +parts-vertex while it runs as a vertex shader
• F3 mobium:modules and mobium:timings keep their lines together in one column, and the static cache's long line is split into three, so they no longer overlap the other column