Windows support for GPULlama3.java

[stratika]

This PR provides support and instructions to install and run on Windows OS, as discussed in #10. The changes included in this PR are listed below:

• README, update the installation part for Windows and removed unnecessary steps.
• Add a set_paths.cmd file for loading environment variables on Windows
• Updated the llama-tornado python program with: i) a fix for the path delimiter in Windows, and ii) loading the variables before launching the execution.

This is the trace after running on an NVIDIA GPU in Windows (no isolation mode):

python llama-tornado --gpu --verbose-init --opencl --model beehive-llama-3.2-1b-instruct-fp16.gguf --prompt "Tell me a joke"
WARNING: Using incubator modules: jdk.incubator.vector
Parse beehive-llama-3.2-1b-instruct-fp16.gguf: 1427 millis
Loading model weights in TornadoVM format (loading F16 -> F16)
Load LlaMa model: 2911 millis
Starting TornadoVM initialization...
TornadoVM GPU execution plan creation: 3117.03 ms
Java to GPU JIT compiler warmup: 3383.45 ms
Transfer read-only weights to GPU: 1043.90 ms
Finished TornadoVM initialization...

I'll try to come up with a clean one. Here's a joke: 

Why was the computer and the keyboard and the mouse all invited to the wedding? 

Because they were all already married. 

achieved tok/s: 19.62. Tokens: 54, seconds: 2.75

[mikepapadim]

LGTM, thanks