Lift hook command note injection into `SkillsIntegration` base class

[mnriem]

Problem

Hook commands in extensions.yml use dotted IDs (e.g., speckit.git.commit) but skills use hyphenated names (speckit-git-commit). Without an explicit note in the generated SKILL.md, agents emit /speckit.git.commit which doesn't resolve.

Claude (#2346) and Codex (#2503) both solve this by independently overriding setup() and post_process_skill_content() with identical _HOOK_COMMAND_NOTE constants and _inject_hook_command_note regex logic. Copilot (skills mode) and Vibe have the same latent bug — they don't inject the note.

Prerequisites

This should land after #2503 is merged, since it will deduplicate and lift the hook note logic that #2503 adds for Codex.

Proposed changes

1. Move the post-processing loop into SkillsIntegration.setup() so it calls self.post_process_skill_content() after writing each SKILL.md. This eliminates the duplicated loop in Claude, Codex, Copilot, and Vibe.

2. Move _inject_hook_command_note into SkillsIntegration and call it from the base post_process_skill_content(). The dot-to-hyphen translation is inherent to how skills map hook IDs — it's not agent-specific.

3. Remove the duplicated logic from ClaudeIntegration and CodexIntegration. Claude's post_process_skill_content() would still override for its agent-specific flags (user-invocable, disable-model-invocation) but call super() to get the hook note.

Affected integrations

Integration	Has post-process loop	Has hook note	After refactor
Base (SkillsIntegration)	No	No	Both
Claude	Yes (duplicated)	Yes	Simplified
Codex	Yes (duplicated, #2503)	Yes	Simplified
Copilot (skills)	Yes (duplicated)	No — bug	Fixed
Vibe	Yes (duplicated)	No — bug	Fixed
Agy	No	No	Fixed
Devin	No	No	Fixed