feat(mosaic): drill-down main menu + provider-first flow + quick start

Replace the linear 12-stage wizard interrogation with a menu-driven
architecture. Key changes:

- Main menu with drill-down sections: Quick Start, Providers, Agent
  Identity, Skills, Gateway, Advanced, Finish & Apply
- Quick Start path: 3 questions (API key, admin email, password) with
  sensible defaults for everything else. Target: under 90 seconds.
- Provider-first flow: LLM API key collection moves to the front of
  the wizard, with auto-detection of Anthropic vs OpenAI from key prefix
- Deterministic intent/naming: user picks an intent category (general,
  software-dev, devops, research, content, custom) and gets a thematic
  agent name proposal (Mosaic, Forge, Sentinel, Atlas, Muse)
- Headless backward compat: MOSAIC_ASSUME_YES=1 + env vars still work
  unchanged, with new MOSAIC_AGENT_NAME and MOSAIC_AGENT_INTENT vars
- Version bump: 0.0.26 -> 0.0.27

Closes #438

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

packages/mosaic/src/constants.ts

@@ -26,6 +26,53 @@ export const DEFAULTS = {
 | (add your git providers here) | | | |`,
 };
 
+/** Preset intent categories with display labels and suggested agent names. */
+export const INTENT_PRESETS: Record<
+  string,
+  { label: string; hint: string; suggestedName: string }
+> = {
+  general: {
+    label: 'General purpose assistant',
+    hint: 'Versatile helper for any task',
+    suggestedName: 'Mosaic',
+  },
+  'software-dev': {
+    label: 'Software development',
+    hint: 'Coding, debugging, architecture',
+    suggestedName: 'Forge',
+  },
+  devops: {
+    label: 'DevOps & infrastructure',
+    hint: 'CI/CD, containers, monitoring',
+    suggestedName: 'Sentinel',
+  },
+  research: {
+    label: 'Research & analysis',
+    hint: 'Data analysis, literature review',
+    suggestedName: 'Atlas',
+  },
+  content: {
+    label: 'Content & writing',
+    hint: 'Documentation, copywriting, editing',
+    suggestedName: 'Muse',
+  },
+  custom: {
+    label: 'Custom',
+    hint: 'Describe your own use case',
+    suggestedName: 'Mosaic',
+  },
+};
+
+/**
+ * Detect LLM provider type from an API key prefix.
+ */
+export function detectProviderType(key: string): 'anthropic' | 'openai' | 'none' {
+  if (!key) return 'none';
+  if (key.startsWith('sk-ant-')) return 'anthropic';
+  if (key.startsWith('sk-')) return 'openai';
+  return 'none';
+}
+
 export const RECOMMENDED_SKILLS = new Set([
   'brainstorming',
   'code-review-excellence',