Improved and fixed template for GPT-OSS (again). Includes preserve_thinking and fix for Unsloth-induced bug

I posted an updated GPT-OSS template a couple of months ago, which was based on Unsloth's version. It turns out that both Unsloth's version (and, thus, mine) contain a very serious bug that can degrade the model when chat history is replayed and contains previous reasoning (aka the analysis channel)