[Performance]: Cache the OpenRouter model catalog with a TTL

[shauryagangrade]

Problem

list_free_models hits https://openrouter.ai/api/v1/models on every call.
/models and startup both fetch the full catalog each time — a wasted
network round-trip and a slow menu on a flaky connection.

Proposed approach

• Add a short-lived in-process cache (e.g. 10-60 min TTL) so repeated calls
  within a session reuse the last response.
• Keep a 0-TTL escape hatch for anything that needs fresh data.
• Respect the existing best-effort contract (empty list + error on failure).

Where to look

• gcode/models.py:18-55 (list_free_models).

Acceptance criteria

• Two consecutive list_free_models() calls inside the TTL make one HTTP request.
• Cache invalidation on expiry is tested.

Difficulty

Low — great first issue.

[aryansk]

Working on this — adding an in-process TTL cache to list_free_models (30-min default, ttl_seconds=0 escape hatch, failed fetches never cached).