#1995 Request: Add Q4_1 PLE Dequantization Support

Description

I am running Strata on Windows with an AMD Radeon RX 9070 (16 GB VRAM), 64 GB of system RAM, and the HIP backend. This setup makes it possible to run certain 4-bit quantized MoE models locally, even with relatively limited GPU memory.

While testing the [[Qwen3.8 Flash-Next AD IQ4_XS M64 model](https://huggingface.co/AtomicChat/Qwen3.8-Flash-Next-GGUF)](https://huggingface.co/AtomicChat/Qwen3.8-Flash-Next-GGUF), I discovered that its `per_layer_token_embd.weight` tensor uses Q4_1 quantization and requires Q4_1 PLE dequantization support in Strata.

With AI-assisted development, I implemented the missing Q4_1 PLE dequantization path and tested it against the actual GGUF data. The validation passed with bit-identical results.

Would you consider integrating Q4_1 PLE dequantization support into Strata upstream?

Full official M64 support would be welcome, but it is not a requirement for this request. Adding Q4_1 PLE support alone would already make the engine more compatible with models that use this format and would save users from maintaining custom patches across releases.

I can provide the patch and validation details if helpful.

Thank you for your work on Strata!