feat: Agent Shield integration - 5-stage guardrails engine for AGT (#1805)

* docs: add architecture and build/test/lint sections to copilot-instructions.md

Add high-level architecture overview (package map, SDK paths) and
build/test/lint commands for all languages (Python, TypeScript, Rust,
.NET, MCP server, docs) including single-test invocations.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* feat: add Agent Shield integration adapter as guardrails engine

Integrate Microsoft Agent Shield as AGT's per-call guardrails engine,
providing 5-stage validation (input, state, tool execution, post-tool,
output) while AGT retains identity, trust scoring, audit, and lifecycle.

- AgentShieldKernel wraps Agent Shield runtime/session lifecycle
- Protocol-based interface (no hard SDK dependency)
- Trust score injection as Agent Shield variables for policy gating
- Fail-closed error handling with configurable fail-open mode
- Mock runtime for testing without Agent Shield SDK installed
- Full observability: per-stage history, stats, violation callbacks
- 40 tests covering all stages, blocking, errors, redaction, lifecycle

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: imran-siddique