docs(lessons-learned): record ZeroGPU workshop load-capacity results

docs: full rewrite for Transformers + ZeroGPU backend

docs: move investigation docs to docs/archive/

docs: add lessons-learned.md — consolidated dead-ends reference

docs: add vLLM backend replacement implementation plan

docs: add vLLM backend replacement design

docs: add vLLM throughput spike design

docs: move durable performance records out of gitignored superpowers/

Revert quantization code — disproven by profiling (GPU idle at 17%)

docs: fill in profiling findings from T4 diagnostic runs

docs: add profiling run methodology and findings template

docs: add request-lifecycle profiling implementation plan

docs: add request-lifecycle profiling design spec

docs: default to Gemma 3 1B, document MODEL_SIZE switching

docs: add stop button design spec

docs: update AGENTS.md, README.md, and architecture review to reflect real LLM backend

feat: add frontend error handling — input validation, backend error detection, disabled buttons on load

docs: add architecture review and refine AGENTS.md

Clean up repo

feat: complete multi-state token probability display feature

docs: add design for multi-state token probability display feature

docs: add design spec and implementation plan for LLM workshop Gradio app