fix(models): merge curated list with live /v1/models for zai provider

[x7peeps]

Problem

When using the Z.AI (GLM) provider, the /model picker shows only models returned by the live /v1/models endpoint, silently dropping entries from the curated _PROVIDER_MODELS list. When the live endpoint omits newer models (e.g. glm-5.2), users can't see them in the picker even though they work when set directly via /model or config.

Root Cause

In provider_model_ids() (hermes_cli/models.py:2379), the generic profile-based live fetch path merges curated + live results only for kimi-coding / kimi-coding-cn. The zai provider falls through to return live, discarding the curated list entirely.

Fix

Add zai to the merge set so curated models come first, then live additions are appended — ensuring no curated model is ever hidden by an incomplete live response.

Closes #47162

[kshitijk4poor]

Closing in favor of #46857.

Same root cause as your fix (#46850 / #47162) — the generic provider path returned only the live /models result and dropped the curated _PROVIDER_MODELS list, hiding glm-5.2. Your approach (add zai to the merge set) was correct; #46857 generalizes the same merge to all api-key providers rather than allowlisting one provider at a time, and also fixes the direct-/model-switch validation path.

Thanks for the diagnosis and the link to #47162.