fix: clean up stale Apache 2.0 mentions + restore walkthrough spacing

Post-Fix-B cleanup commit. Three CLAUDE.md spots still referenced
Apache 2.0 (Tech-stack deliverable parenthetical, file-tree comment,
and the agent's verifier-compatibility tweak in walkthrough.md).
Restored the idiomatic blank line under '## License' in walkthrough.md
since the original verifier bug is irrelevant going forward.

Historical Progress-log entry from 2026-05-07 left intact (it accurately
describes the state of the project on that date).

CLAUDE.md

@@ -157,7 +157,7 @@ Win the AMD Developer Hackathon (LabLab.ai, May 2026), Track 3, with a real-time
 ## Tech stack (locked)
 
 - Languages: Python 3.12 (primary)
-- Submission deliverable: Hugging Face Space (Gradio app, public, Apache 2.0)
+- Submission deliverable: Hugging Face Space (Gradio app, public, MIT)
 - Inference backend: FastAPI on AMD Developer Cloud (single MI300X instance), exposed as OpenAI-compatible API
 - Transport: HTTPS for V1; WebSocket only if latency demands it post-Day-2
 - Pipeline (concurrent on one MI300X):
@@ -204,7 +204,7 @@ git push huggingface main
 ```
 /Users/lucaslt/Documents/side-gig/amd-hackathon/
 ├── README.md                       # HF Space card via frontmatter
-├── LICENSE                         # Apache 2.0
+├── LICENSE                         # MIT
 ├── CLAUDE.md
 ├── .claude/
 ├── requirements.txt

docs/walkthrough.md

@@ -92,4 +92,5 @@ The single-GPU concurrency story is the AMD pitch. This V1 fits on H100;
 the architecture has clear headroom on MI300X for higher-quality V2 models.
 
 ## License
+
 MIT.