Optimize koboldcpp.sh for local users

[henk717]

Right now koboldcpp.sh is to heavy handed for local users, this didn't matter when I made it but with how large llamacpp has gotten the builds get really long for what they don't need.

At the same time people may not have CUDA 12.1+, ARM, AMD, etc.

These changes make it easier to run from source without knowing what you are doing:

• Portable builds now require KCPP_PORTABLE so people don't end up compiling what they don't need by default.
• CUDA 11.4 / ROCm is automatically selected when its detected that this is required and not manually specified
• ARM mamba is now automatically used if its an ARM system, systems that aren't x64 or ARM are now rejected.
• ARM binaries say so in the name.

[henk717]

Still need to verify the end result binaries but ran out of time.
They are all running in my branch overnight so you have a chance to verify them yourself if you want to.

[henk717]

CI failed, will look at it another time

[henk717]

Ended up fixing those, back to reviewable. Builds are running and can confirm locally it spits out the required files of the bin.

[LostRuins]

still draft so i guess its still WIP?

[henk717]

It mostly needs the binaries validated but I did do last minute fixes last night.

[henk717]

AMD binary had issues so on hold until I validate everything properly