feat(nd): Brain Dump → Task Chunker — AI micro-task generator with effort scoring

[welshDog]

🧠 What This PR Does

Adds the Brain Dump → Task Chunker — audit Feature #4 from the neurodivergent IDE roadmap.

Paste your chaotic brain dump — tasks, worries, half-formed ideas, all of it — and get back a clean, prioritised list of micro-tasks with effort scores and a single "Start Here" recommendation.

💬 "As an ADHD developer with 40 tabs open and a head full of noise, I need to externalise my brain and get ONE clear next action."

---

📦 Files Added

File	Purpose
lib/brainDump.ts	Chunking engine — parser, categoriser, effort scorer, focus picker
components/BrainDumpChunker.tsx	UI widget — textarea, results, dismiss chips, ARIA compliant
__tests__/brainDump.test.ts	15 unit tests covering all logic branches

---

⚡ How It Works

Raw brain dump text
  → Split on: commas, newlines, semicolons, periods, "and"
  → Each fragment → detect category + urgency + effort score
  → Sort: urgent first → lowest effort first
  → Output: task chips + "Start Here" suggestion + overload warning

---

🎨 Features

• 7 task categories with emoji anchors: 💻 code 🔍 research 💬 comms 📋 admin 🎨 creative 🚧 blocked 📌 other
• Fibonacci-lite effort scores: ⚡ Quick → 🟢 Easy → 🟡 Medium → 🔴 Big → 🦅 Epic
• ⚠️ Overload warning fires when >7 tasks detected
• 🎯 "Start Here" — picks urgent+lowest-effort task automatically
• ✕ Dismiss chips — swipe tasks away as you do them
• aria-live="assertive" — announces result count + suggested focus to screen readers
• role="alert" on overload warning — interrupts immediately
• Zero dependencies — pure TypeScript, no AI API call needed (heuristic engine)

---

🧪 Test Coverage (15 tests)

# All pass locally alongside existing 26 tests
- Empty / whitespace input → empty result
- Comma + newline splitting
- Code / comms category detection
- Urgent flag detection
- overloadWarning threshold (>7 tasks)
- suggestedFocus picks urgent first, then lowest effort
- Task text capitalisation
- EFFORT_LABELS + CATEGORY_EMOJI completeness guards

---

🧠 ND Design Rationale

• No AI API call needed — heuristic engine = instant response, works offline, zero latency anxiety
• One "Start Here" task — eliminates the paralysis of choosing from a list
• Dismiss chips — visual progress feedback = dopamine reward loop 🔥
• Overload warning — prevents the "I'll just add one more thing" spiral
• Plain textarea — no special syntax to learn, just write

---

🧠 Audit Feature #4 — HyperCode V2.0 Neurodivergent IDE remediation plan.