fix: wait for model catalog before opening picker

[Michaelyklam]

Thinking Path

• The bottom model picker is backed by a hidden native <select> plus a visible custom dropdown.
• /api/models can correctly return OpenAI Codex models while the visible dropdown still renders the static HTML fallback if it opens before async model hydration finishes.
• The fix should preserve non-blocking page boot, but prevent the visible picker from rendering stale options.
• The safest narrow change is to await the already-exposed window._modelDropdownReady before first render, and re-render if hydration completes while the picker is already open.

What Changed

• Makes toggleModelDropdown() async and waits for window._modelDropdownReady before calling renderModelDropdown().
• Re-renders the visible custom dropdown after populateModelDropdown() replaces the hidden <select> when the picker is already open.
• Adds regression coverage for the [bug] Model picker can show stale static options and hide Codex models on fast open #1743 race.
• Updates an existing source-boundary regression to accept the now-async toggle function.

Why It Matters

• Fixes a race where fast-open users could see stale static OpenAI/Anthropic/Other options and miss the configured OpenAI Codex models.
• Keeps backend/catalog behavior unchanged; this is strictly a frontend synchronization fix.

Fixes #1743.

Verification

• RED: tests/test_issue1743_model_picker_race.py failed before implementation with missing async wait/re-render assertions.
• Targeted: /home/michael/.hermes/hermes-agent/venv/bin/python -m pytest tests/test_issue1743_model_picker_race.py tests/test_ollama_model_chip_label_regression.py tests/test_byok_model_dropdown.py tests/test_model_picker_badges.py tests/test_issue1567_nous_picker_capacity_and_symmetry.py -q → 48 passed in 11.33s
• Full suite: env -u HERMES_CONFIG_PATH -u HERMES_WEBUI_HOST /home/michael/.hermes/hermes-agent/venv/bin/python -m pytest tests/ -q → 4598 passed, 2 skipped, 3 xpassed, 1 warning, 8 subtests passed in 382.74s
• git diff --check passed.
• Browser QA on an isolated local WebUI server:
  • Confirmed toggleModelDropdown() waits on a delayed window._modelDropdownReady before opening.
  • Confirmed a delayed catalog update renders OpenAI Codex rows instead of the stale static fallback.
  • Confirmed the hydrated dropdown includes Configured, Gemini, OpenAI Codex, and OpenRouter groups with Codex rows such as GPT 5.5, GPT 5.4, GPT 5.3 Codex, and GPT 5.3 Codex Spark.

Risks / Follow-ups

• Waiting for window._modelDropdownReady can delay first dropdown open by the /api/models request duration, but avoids showing an incorrect catalog. If /api/models fails, the existing promise catches and the static fallback remains available.
• No backend model catalog behavior is changed.
• No screenshots included because this is an interaction/timing race with no persistent visual redesign; browser QA was DOM/behavior verified on an isolated server.

Model Used

• OpenAI Codex / gpt-5.5
• Tool use: file inspection, pytest, isolated local WebUI browser QA, git/GitHub CLI.