Bootstrap scoped agent knowledge for the SDK

[mthalman]

Problem

Agent guidance in dotnet/sdk is spread across root instructions, area-specific files, skills, and contributor documentation. Repository-wide context is repeatedly embedded in always-loaded instructions, making it expensive to load, difficult to navigate, and prone to becoming inconsistent as the repository evolves.

There is also no defined ownership or maintenance workflow for durable agent knowledge.

Related to #55185

Fix

Introduce a scoped knowledge base under .github/memory/ with an index that tells agents what context exists and when to load it.

Repository-wide architecture, file locations, public surfaces, conventions, testing strategy, and durable gotchas now have focused canonical documents. Root Copilot instructions retain only always-needed orientation and guardrails, linking to detailed context on demand.

An update-docs skill defines documentation ownership and ensures relevant memory, instructions, skills, agent definitions, and contributor documentation are reviewed after code changes. Code-review guidance now checks these artifacts for drift.

Dependency

This change assumes #55787 merges first and uses the run-tests contract introduced there.

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).
2 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

Pull request overview

Adds a new scoped knowledge base under .github/memory/ (with an index + focused canonical docs) and refactors the root Copilot instructions and review guidance to reference it, alongside introducing an update-docs skill to keep AI/contributor artifacts in sync after code changes.

Changes:

• Introduces .github/memory/ with an INDEX.md loading map plus focused docs for architecture, API surfaces, conventions, file map, testing strategy, and known issues.
• Adds an update-docs skill that defines a maintenance workflow/checklist for memory + other contributor/agent artifacts.
• Updates .github/copilot-instructions.md and the code-review skill to point at the new memory system (including new references to run-tests).

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
.github/skills/update-docs/SKILL.md	Adds a documentation/memory maintenance checklist skill.
.github/skills/code-review/SKILL.md	Expands “stale AI context” review guidance to include .github/memory/*.md and .claude skills.
.github/memory/ARCHITECTURE.md	Adds a canonical architecture/ownership/flow map.
.github/memory/API_MAP.md	Adds a canonical map of user-facing and extension surfaces and where they’re defined.
.github/memory/CONVENTIONS.md	Adds a canonical conventions/dependency/framework constraints summary with pointers to owners.
.github/memory/FILE_MAP.md	Adds a navigation/ownership-oriented directory map for the repo.
.github/memory/INDEX.md	Adds the “read first” loading map for the memory knowledge base.
.github/memory/KNOWN_ISSUES.md	Adds durable gotchas/workarounds for common repo pitfalls.
.github/memory/TESTING_STRATEGY.md	Adds test strategy + canonical sources map for testing guidance ownership.
.github/copilot-instructions.md	Refactors root instructions to defer details to memory docs and reference run-tests/incremental-test.