chore(#1805): replace hardcoded model enums with authoritative refs

[fullsend-ai-coder]

The pr-review SKILL.md hardcoded specific model names (opus, sonnet, haiku, claude-sonnet-4-6) when describing valid values for the Agent tool's model parameter. When upstream tool schemas expanded, these hardcoded lists became stale and caused false positive api-contract findings in review runs.

Replace the two prose sections that enumerate model names with generic references to the Agent tool's model parameter definition. The sub-agent table still shows current frontmatter values (which is factual usage, not an exhaustive valid-value list). The sub-agent frontmatter files themselves are unchanged since they contain actual configuration, not documentation of allowed values.

Note: make lint could not run due to sandbox Go toolchain permission issue (unrelated to this change). This is a docs-only change with no Go code modifications.

---

Closes #1805

Post-script verification

• Branch is not main/master (agent/1805-remove-hardcoded-model-enums)
• Secret scan passed (gitleaks — de4ce7110e460377f26ea2ed285d5e8c3f4a560f..HEAD)
• Pre-commit hooks passed (authoritative run on runner)
• Tests ran inside sandbox

[github-actions]

Site preview

Preview: https://ef83d186-site.fullsend-ai.workers.dev

Commit: 7d1c4a5deb7cf9284b7185808df96f2f09077dbc

[fullsend-ai-review]

Review

Findings

No findings.

Two sonnet-tier sub-agents (style-conventions, intent-coherence) could not be dispatched due to model availability on the current Vertex deployment. These are informational gaps only — the correctness sub-agent (opus-tier) completed successfully with no findings, and this is a minimal docs-only change (4 additions, 5 deletions in a single Markdown file) that replaces hardcoded model name enumerations with generic references to the Agent tool's model parameter definition. The change is well-scoped to the linked issue (#1805) and touches no protected paths.