feat: target-aware hook event diagnostics

[danielmeppiel]

Summary

When APM deploys hooks for a package, it should log which hook event names were found and note if they may not be recognized by the target platform. Currently APM silently deploys all events regardless of target.

Background

The hooks specification has no unified standard — three implementations exist with divergent event names and schemas:

Platform	Event naming	Example events	Spec reference
GitHub Copilot Agent (cloud/CLI)	camelCase	sessionStart, preToolUse, postToolUse, userPromptSubmitted	docs.github.com
VS Code (Copilot Chat extension)	PascalCase	SessionStart, PreToolUse, PostToolUse, UserPromptSubmit	code.visualstudio.com
Claude Code	PascalCase	SessionStart, PreToolUse, PostToolUse, UserPromptSubmit, PermissionRequest, FileChanged	code.claude.com

Additionally, each platform supports different events — Claude Code has 25+ events, while GH Agent has 8.

Proposed Behavior

When deploying hooks to a target (e.g., copilot, claude, cursor, codex):

1. Log the event names found in the hook JSON (e.g., "Deployed 3 hook events: preToolUse, postToolUse, sessionStart")
2. Optionally note if events use a naming convention that may not match the target (e.g., "Note: camelCase events detected — these are GitHub Copilot Agent format; VS Code expects PascalCase")

Scope

• Informational only — do NOT block deployment
• Emit via existing logger.info or diagnostics mechanism
• Consider a future --strict mode that warns on unrecognized events per target

Related

• Issue Bug: Hook integrator does not process 'windows' property – scripts not copied and paths not rewritten #520 — original hooks bug (missing windows key)
• The hook command keys (command, bash, powershell, windows, linux, osx) are already handled as a superset; this issue addresses the event name dimension