feat(models): add qwen3.8-max to Nous portal + OpenRouter, replacing qwen3.7-max

[teknium1]

Summary

qwen/qwen3.8-max is now the Qwen entry in both the Nous portal and OpenRouter curated catalogs, replacing qwen/qwen3.7-max (newest-max-replaces-last-max, per the deepseek/kimi convention in these lists).

Changes

• hermes_cli/models.py: OPENROUTER_MODELS + _PROVIDER_MODELS["nous"] — swap qwen/qwen3.7-max → qwen/qwen3.8-max in the Qwen slot.
• agent/model_metadata.py: DEFAULT_CONTEXT_LENGTHS["qwen3.8-max"] = 1_000_000 — without it the slug fuzzy-matched the generic qwen 131K fallback. Verified against OpenRouter live metadata and Nous /v1/models (2026-08-03).
• tests/test_empty_model_fallback.py: swap the incidental catalog fixture to the surviving slug.
• website/static/api/model-catalog.json: regenerated via scripts/build_model_catalog.py.

Deliberately out of scope (scoped rollout)

• _PROVIDER_MODELS["alibaba"] / opencode-go still carry qwen3.7-max — different routes, not named in the request.
• Pricing snapshot skipped: both requested routes bill via official_models_api (live pricing), confirmed with resolve_billing_route.
• Reasoning stale-timeout floor already covered by the qwen3 prefix entry (180s).

Validation

Check	Result
Targeted suites (test_models, test_model_catalog, test_gmi_provider, test_usage_pricing, test_empty_model_fallback, test_model_metadata)	152/152 pass
E2E catalog (temp HERMES_HOME, keys stripped): 3.8-max present, 3.7-max absent, no dupes, correct slot	pass
Context fuzzy match: qwen/qwen3.8-max → 1,000,000; qwen3-max-2026-01-23 snapshot still → 262,144	pass
Billing route nous + openrouter	official_models_api

Infographic

[github-actions]

૮ >ﻌ< ა ci review

_{ran on 835b743}

all good!