Optimize and rewrite texture loading

[Phantomical]

KSPCF's texture loader is pretty fast so normally I wouldn't really bother to optimize it, but I would like to do some additional work here and a refactor + optimize pass will make that easier.

The big changes here are:

• Every texture load is now launched in its own independent coroutine
• Instead of using the normal Texture2D constructor we now use CreateUnititializedTexture2D, as lifted from KSPTL.
• DDS loading is now more async:
  • Create a texture
  • Wait a frame for unity to finish its first upload
  • Use AsyncReadManager to read directly into the array returned by GetRawTextureData()
  • Apply (and make it unreadable if needed)
• The PNG conversion cache is now gone, however we still compress and mark textures unreadable just like it was before.
• PNG/JPG textures are still loaded through UnityWebRequest, normal swizzling is now offloaded to a background thread.
• TGA/MBM/etc. remains mostly the same, though normal swizzling is offloaded.

I do have plans to reintroduce a cache in a different form, so this PR keeps the popup around for that use case.

Despite the fact that we are now compressing all the png textures every time, we still get a decent ~25% improvement compared to the old KSPCF texture loader.

Here's what the baseline looks like (2nd reload, warm cache, KSPCF 1.40.1 off CKAN)

[LOG 20:18:54.796] [KSPCF:FastLoader] 13th Gen Intel(R) Core(TM) i7-13700H | 65208 MB | NVIDIA RTX A1000 6GB Laptop GPU (6001 MB)
Total loading time to main menu : 97.088s
- Configs and assemblies loaded in 10.006s
- Configs reload done in 6.669s
- Configs translated in 0.075s
- 13942 assets loaded in 24.458s :
  - 243 audio assets (144.512 MiB) in 4.123s, 35.053 MiB/s
  - 8240 texture assets (7.569 GiB) in 10.593s, 731.758 MiB/s
  - 5430 model assets (1.586 GiB) in 9.740s, 166.818 MiB/s
- Asset bundles loaded in 0.555s
- GameDatabase (configs, resources, traits, upgrades...) loaded in 0.427s
- Built-in parts copied in 0.044s
- Part and internal configs extracted in 0.010s
- 3694 parts and 20662 modules compiled in 28.029s
  - 5.6 modules/part, 7.588 ms/part, 1.357 ms/module
  - PartIcon compilation : 10.499s
- 161 internal spaces and 523 props compiled in 0.938s
- 2 DLC (Making History, Breaking Ground) loaded in 0.387s
- Planetary system loaded in 10.382s

And here's with this PR applied

[LOG 02:04:25.475] [KSPCF:FastLoader] 13th Gen Intel(R) Core(TM) i7-13700H | 65208 MB | NVIDIA RTX A1000 6GB Laptop GPU (6001 MB)
Total loading time to main menu : 89.487s
- Configs and assemblies loaded in 11.225s
- Configs reload done in 5.646s
- Configs translated in 0.076s
- 12697 assets loaded in 17.452s :
  - 243 audio assets (144.512 MiB) in 4.407s, 32.788 MiB/s
  - 7688 texture assets (6.798 GiB) in 7.695s, 904.62 MiB/s
  - 4737 model assets (1.526 GiB) in 5.344s, 292.498 MiB/s
- Asset bundles loaded in 0.493s
- GameDatabase (configs, resources, traits, upgrades...) loaded in 0.337s
- Built-in parts copied in 0.031s
- Part and internal configs extracted in 0.013s
- 3694 parts and 20662 modules compiled in 22.009s
  - 5.6 modules/part, 5.958 ms/part, 1.065 ms/module
  - PartIcon compilation : 8.164s
- 16 internal spaces and 0 props compiled in 0.045s
- 2 DLC (Making History, Breaking Ground) loaded in 0.362s
- Planetary system loaded in 8.935s

This is on a test save with FFT, NFT, BDB, Tantares, SSPX, Heat Control, and all of Sterling Systems - though IVAs are disabled since that seems to be the default?

As a bonus this should perform acceptably well on HDDs since it AsyncReadManager only ever reads one texture at a time.

[Phantomical]

The performance comparisons above are inaccurate. Marking this as draft until I can come up with better ones (and those better ones actually show an improvement)