feat(providers): split zai into 4 plans (Global/China × direct API/Coding Plan) — salvage #13500

[kshitijk4poor]

Summary

Salvage of #13500 by @yuanmingyi — cherry-picked onto current main with authorship preserved.

Replaces the single zai provider (which used fragile runtime HTTP endpoint probing via detect_zai_endpoint) with 4 explicit static providers. Users pick their plan once; credentials resolve deterministically with no startup latency.

Hermes id	Endpoint	Env var	Plan
zai	api.z.ai/api/paas/v4	ZAI_API_KEY	Z.AI (Global direct)
zai-cn	open.bigmodel.cn/api/paas/v4	GLM_API_KEY	Zhipu AI (China direct)
zai-coding-global	api.z.ai/api/coding/paas/v4	ZAI_CODING_API_KEY	Z.AI Coding Plan (Global)
zai-coding-cn	open.bigmodel.cn/api/coding/paas/v4	GLM_CODING_API_KEY	Zhipu AI Coding Plan (China)

Changes

• hermes_cli/auth.py — 4 distinct PROVIDER_REGISTRY entries with per-plan env vars and static base URLs. Drops detect_zai_endpoint / _resolve_zai_base_url (eliminates ~8s probes). zai-cn declared before zai so auto-detect with only GLM_API_KEY prefers China plan.
• hermes_cli/providers.py — HERMES_OVERLAYS gains 3 new entries. Aliases route glm/zhipu → zai-cn; z-ai/z.ai → zai.
• agent/models_dev.py — maps each Hermes id to its models.dev catalog (zhipuai, zai-coding-plan, zhipuai-coding-plan)
• hermes_cli/models.py — curated model lists: direct plans (9 models), coding plans (4 coding-focused models)
• agent/auxiliary_client.py — aux model defaults per plan; vision model for zai/zai-cn
• run_agent.py — _anthropic_preserve_dots() extended to all 4 plans
• trajectory_compressor.py — path-based provider disambiguation (fixes dead code that only compared hostnames)
• All other files: mechanical threading through config, doctor, status, setup, dump, main.py --provider choices, qqbot STT adapter

Follow-up fix

• Added base_url_override to coding plan overlays in providers.py for resilience (matches the pattern used by direct plan overlays rather than relying solely on models.dev)

Compatibility

• provider: zai + only GLM_API_KEY → key resolves via fallback tuple; request hits api.z.ai (may 401 if key was issued for bigmodel.cn — running hermes model re-picks the correct plan)
• provider: zai-cn + GLM_API_KEY → clean
• Auto-detect with only GLM_API_KEY → zai-cn (registry order ensures this)
• ZAI_API_KEY always wins over GLM_API_KEY when both present

Testing

574 tests pass across the affected test files (1 deselected is pre-existing on main):

tests/hermes_cli/test_api_key_providers.py
tests/hermes_cli/test_model_validation.py
tests/hermes_cli/test_model_provider_persistence.py
tests/agent/test_auxiliary_client.py
tests/agent/test_model_metadata.py
tests/agent/test_models_dev.py
tests/hermes_cli/test_model_normalize.py
tests/hermes_cli/test_doctor.py
tests/hermes_cli/test_status.py
tests/gateway/test_qqbot.py

Closes #13500

[yuanmingyi]

Both failing checks look unrelated to this PR — could someone confirm?

• Supply Chain Audit: guessing it flagged hermes_cli/setup.py, but that's the CLI setup wizard module inside the hermes_cli package, not a setuptools install hook — repo packages via pyproject.toml, no root setup.py, and the diff there is purely data. Happy to be corrected if it was something else.
• Tests: the same 8 failures reproduce on latest main (run 24760241591), none touching zai/providers — looks like pre-existing flakes.

[yuanmingyi]

@kshitijk4poor hello, can you please tell what is blocking in detail or help to bypass the failing checks?

Both failing checks look unrelated to this PR — could someone confirm?

• Supply Chain Audit: guessing it flagged hermes_cli/setup.py, but that's the CLI setup wizard module inside the hermes_cli package, not a setuptools install hook — repo packages via pyproject.toml, no root setup.py, and the diff there is purely data. Happy to be corrected if it was something else.
• Tests: the same 8 failures reproduce on latest main (run 24760241591), none touching zai/providers — looks like pre-existing flakes.

[teknium1]

Dont worry about the failing tests, and the supply chain audit is just for us to take more caution when reviewing

[yuanmingyi]

I'll discard this PR and open a new PR with simpler changes (just one new zai coding plan provider added) at #34323 @teknium1 @alt-glitch