LM Studio + Zoo + Qwen 3.6 issues

I'm currently running an Unsloth quant of Qwen3.6-35B-A3B and I've managed to speed up my output to ~80 tk/s by offloading all experts to CPU. I'm on a Legion 7i laptop, 5080 with 16 GB VRAM. I mainly use the models for Zoo (formerly Roo) integration into VSCode. I'm running into two issues that I'm not sure how to fi, and I can't seem to find anything online.

1. Somewhere in the LM Studio to Zoo pipeline, my character encoding is mismatched, and so when the model attempts to write a comment with, say, an em-dash, it instead comes out as a few random characters.

2. The more pressing issue is that when it tries to edit code, it fails on the basis of not having a 100% match. The issue is, the code exactly matches what it aimed to grab, however it ALSO grabbed the few surrounding lines.

If anybody has any advice for this, that would be wonderful. Thanks!