Resync model metadata with current provider lineups

[grikomsn]

Summary

Catalog enrichment resync against today's live /models lists and models.dev:

• deepseek-flash appears in authenticated Go discovery but has no models.dev opencode-go entry; it is fully cataloged upstream under DeepSeek (V4.1 Flash, 1M context, 393K output, image input, low/high/max reasoning efforts, real costs $0.15/$0.60/$0.003). Mirrored into SUPPLEMENTAL_MODELS so authenticated users get full metadata. Routes chat-completions (npm @ai-sdk/openai-compatible).
• Mirror canonical Zen-provider entries for anonymous-discovery-only Go ids: minimax-m2.5, kimi-k2.5, glm-5.1, glm-5, and qwen3.5-plus (keeps its @ai-sdk/anthropic npm → Messages endpoint). Mirrors only carry data a canonical upstream entry already publishes for the same model; nothing guessed.
• Filter internal OpenCode smoke-test ids (test, test-novita-dsf4.1, …) that leak into authenticated Zen discovery instead of listing them as unenriched picker entries.
• No routing changes: resolveEndpointKind already resolves every new id correctly.
• No fallback-list changes: all current fallback ids are still live and accurate.

jev-1.13 / jev-1.13-free (Zen) and omen-alpha, mimo-v2-pro / mimo-v2-omni (Go anonymous) remain unenriched — no upstream signal anywhere; they will be enriched once models.dev catalogs them (supplemental entries are superseded by canonical ones automatically).

Verification

• npm test passes (new supplemental-mirror and test-id filter tests).
• Live lists fetched auth + anonymous for both modes; models.dev cross-checked.
• Changeset included (patch, package opencode-bridge-copilot-chat).