Add the judge video and Devpost submission kit

[CodeBuildder]

Closes #124

What changed

• adds SUBMISSION_KIT.md as the single recording and submission source
• provides a frame-by-frame 1:55 judge video plan with exact screen, click, narration, and timing
• presents every shot as separate SHOW, DO BEFORE SPEAKING, DO WHILE SPEAKING, SAY, and END ON cues
• recommends recording the screen first and adding narration as a voice-over
• uses conversational, first-person narration instead of formal product-copy language
• centers the differentiator: early kernel/runtime/network containment plus proactive resilience testing before customer impact
• aligns the recording with the official under-three-minute rule and required Codex/GPT-5.6 coverage
• adds dedicated GPT-5.6-in-product and Codex-build-workflow scenes
• frames the USP explicitly as layered autonomous protection across admission, kernel/runtime, L3/L4/L7 network enforcement, service resilience, and Sentinel coordination
• validates the spoken script at 364 words across a paced 2:55 cut
• documents truthful claim boundaries for replayed, simulator, observed, and live-Chaos evidence
• includes recording fallbacks so a transient API, UI, or cluster problem does not derail the presentation
• defines the required polished screenshot set and recommended Devpost cover image
• provides copy-ready Devpost sections covering inspiration, product, architecture, OpenAI usage, differentiation, challenges, accomplishments, and next steps
• adds a final recording and submission quality gate
• records Phases 15 and 16 in BUILD_WEEK.md
• links the kit prominently from the root README

Why

The product is feature-complete enough for submission, but judge impact now depends on a concise, rehearsable, and technically truthful story. This change turns the platform into a repeatable two-minute presentation without requiring the presenter to improvise or overstate synthetic evidence.

Judge impact

Every product claim is paired with a visible screen or evidence artifact. OpenAI reasoning, the Sentinel Operations Graph, human governance, provenance, and verified recovery are all introduced within the first 90 seconds.

Validation

• git diff --check
• narration word count: 364 words
• planned final runtime: 2:55 with per-shot conversational pacing
• official rule checked: video must be under three minutes with audio covering what was built and how Codex and GPT-5.6 were used