Docs/agent posture integrations (DO NOT DELETE) - #54

Adds a new Agent Posture section to the docs with client-facing setup
guides for all four supported integrations. Covers required permissions,
step-by-step setup, feature availability tables, known limitations, and
troubleshooting for each provider.
- agent-posture/overview.mdx: product overview and integration summary
- agent-posture/azure.mdx: Azure AI Foundry (v2) + OpenAI Classic (v1)
- agent-posture/gcp-vertex-ai.mdx: GCP Vertex AI Reasoning Engines
- agent-posture/mistral.mdx: Mistral AI agents and document libraries
- agent-posture/m365-copilot.mdx: M365 Copilot and Copilot Studio bots
- docs.json: adds Agent Posture dropdown to navigation
Closes #60, #82, #94, #116 

Reorders the navigation dropdown so Agent Posture appears between
Trustgate and TrustTest, reflecting its position in the product suite.

Adds self-hosted SVG icons for all four Agent Posture integrations:
- azure.svg (Microsoft Azure, #0078D4)
- googlecloud.svg (Google Cloud, #4285F4)
- mistralai.svg (Mistral AI, #FF7000)
- m365.svg (Microsoft Copilot Studio — official icon from adoption.microsoft.com)
Used both as page icons in frontmatter (sidebar) and in the
brand identity card rendered at the top of each integration page.

- Fix Classic v1 usage source: "Via Azure Monitor" -> "Via Assistants API (Threads & Runs)"
- Fix Azure AI Foundry v2 usage source: "Via Log Analytics" -> "Via Application Insights"
- Extend auth sentence to include Mistral API-key model alongside
 service principals and service accounts

Accuracy fixes (from internal review):
Azure:
- Remove incorrect "No tool call telemetry" limitation; App Insights
 tracks code_interpreter, file_search and function calls via gen_ai spans
- Rewrite telemetry section: source is Application Insights
 (AppDependencies), not Diagnostic Settings/Log Analytics alone
- Split feature matrix into separate v2 (Foundry) and v1 (Classic)
 tables; v1 usage is API-based and needs no Log Analytics Reader
- Clarify "conversation discovery" as telemetry aggregation signals,
 not a list endpoint
GCP:
- Replace "comma-separated regions" with correct discover_all /
 selected_locations / legacy location model, including JSON examples
Mistral:
- Soften content safety claim to heuristic advisory; clarify
 safe_prompt as the only verifiable runtime signal
M365 Copilot:
- Mark AgentInstance.Read.All as optional; document graceful
 degradation (403/404 skips registry, Dataverse continues)
- Add "Usage is all-time only" known limitation
- Remove misleading AgentInstance.Read.All bullet from troubleshooting
Visual improvements:
- Replace all emoji indicators (✅❌⚠️) with styled Yes/No/Warning spans
- Convert setup flows to <Steps> / <Tabs> (CLI + Portal variants)
- Convert troubleshooting sections to <AccordionGroup> / <Accordion>
- Add branded identity card (logo + subtitle) at top of each page
- Add icon frontmatter on each page for sidebar display
- Remove redundant hero descriptions (now covered by brand card)

Covers all four Azure agent infrastructure models:
- Azure AI Foundry (v2) — unchanged
- AI Hub (ML Workspace-backed, kind=Project) — new section
- Azure OpenAI Classic (v1) — unchanged
- Legacy ML Workspaces (kind=Default) — new section
Per-model changes:
- Discovery tables for AI Hub and legacy ML Workspace
- Updated permissions section with model-specific RBAC roles:
 - AI Hub: Reader (sub) + Azure AI Developer (Hub/Project)
 - Legacy ML: Reader (sub) + Azure Machine Learning Data Scientist
- Corrected the Azure AI Developer warning: it IS needed for Hub model
- Added note on how to identify which model a project uses

...tion-level RBAC setup
- Add Monitoring Reader role to telemetry permissions (was missing despite being required by azure_metrics.py for Hub-backed agent metrics via Azure Monitor)
- Add a new consolidated section "All infrastructure types — single subscription setup" with a tiered role table and a single CLI block covering every infrastructure model
- Expand Step 3 (optional telemetry) with subscription-scope and resource-scope tabs for both Log Analytics Reader and Monitoring Reader
- Add AI Hub feature availability matrix gated on Monitoring Reader
- Expand telemetry source summary table to include Hub metrics row
- Add troubleshooting entry for Hub metrics showing zero (missing Monitoring Reader)
- Note that both Log Analytics Reader and Monitoring Reader can be assigned at subscription level to cascade automatically

...missions
- Add agent-posture/github.mdx: full GitHub App setup guide covering
 authentication, required permissions (contents:read, metadata:read),
 installation steps, scanned resources, and security considerations
- Register github.mdx in docs.json navigation under agent-posture group
- Update m365-copilot.mdx: add AiEnterpriseInteraction.Read.All (Graph
 application permission) and ActivityFeed.Read (Office 365 Management API,
 scope https://manage.office.com/.default) to the optional permissions table
 and Azure CLI setup steps — both are used in code but were undocumented
- Update gcp-vertex-ai.mdx: add note that cloudasset.googleapis.com is an
 optional API that accelerates location discovery (requires
 roles/cloudasset.viewer) but the connector falls back to parallel regional
 probing without it

...covery
Replace the outdated Mistral "advisory warning" content safety section with
accurate documentation of native guardrails discovery: the moderation_llm_v2
policy structure, all v2 category keys, and the guardrails JSON schema.
Add a full Model Armor section to the GCP Vertex AI page covering:
- modelarmor.googleapis.com API requirement
- roles/modelarmor.viewer + custom modelArmorFloorReader role (with commands)
- Template and floor setting schema with field-level descriptions
- All filterType and confidenceLevel enum values
- A complete example guardrails JSON object
- Partial access behavior (templates vs floor setting read independently)
- Feature availability table rows for both guardrails features
- Two new troubleshooting accordions for missing guardrails scenarios