fix: consume per-route reasoning mode capability

[erudenko]

Summary

• consume Models Index aggregators[].reasoning.mode from the exact selected serving route
• enable pro injection only when the route is supported and its native values contain pro
• treat rejected and unknown as hard no-injection states, even if a legacy sibling preset row exists
• retain same-provider routeVariant lookup only when an older cache has no typed route fact
• preserve explicit --model-params precedence over catalog-driven injection

This completes the client side of Models Index PR #47.

Verification

• focused route/preset and wire-level suite: 22 passed
• full safe suite: 3,022 CLI tests + 20 bridge tests passed, 17 live-only tests skipped, 0 failed
• monorepo type-check passed
• monorepo lint passed (existing warning-only debt remains)
• CLI and macOS bridge production builds passed
• production Models Index readback currently reports OpenAI/OpenRouter supported, OpenAI Codex rejected, and OpenCode Zen unknown for the applicable GPT-5.6 routes

[claude]

⚠️ Code review skipped — your organization's overage spend limit has been reached.

Code review is billed via overage credits. To resume reviews, an organization admin can raise the monthly limit at claude.ai/admin-settings/claude-code.

Once credits are available, push a new commit or reopen this pull request to trigger a review.