fix(ci): unblock main Tests workflow — 6 unrelated failures across aux model, acp version, gateway test mocks, BUILTIN_SUBCOMMANDS, post_setup parametrize

[teknium1]

Summary

Unblocks main Tests workflow that has been failing for hours. Six logical commits, each scoped to one failure class.

Root causes

1. agent/auxiliary_client.py — _OPENROUTER_MODEL was set to google/gemini-2.5-flash but the intended default is google/gemini-3-flash-preview (matches _NOUS_MODEL). Production constant fixed; tests already asserted the correct name.
2. acp_registry/agent.json — manifest still pinned to 0.13.0 while pyproject.toml has bumped to 0.14.0. Updated manifest version + uvx package pin.
3. tests/gateway/test_teams.py + test_google_chat.py — _install_fake_aiohttp lambdas don't accept **kwargs, so the production trust_env=True arg from Batch salvage group 4: 9 low-risk new-contributor PRs (proxy-env/gateway-fixes/security-headers/custom-providers) #27308's parity fix raised TypeError. Added **kwargs to both fake-session factories.
4. tests/gateway/test_{background_command,telegram_thread_fallback,voice_command}.py — production now adds direct_messages_topic_id to Telegram DM metadata alongside the legacy telegram_dm_topic_reply_fallback flag (so synthetic/queued messages route to the right topic). Three test assertions updated to include the new key.
5. hermes_cli/main.py — _BUILTIN_SUBCOMMANDS was missing send. The hermes send CLI subcommand is live but not in the gating set; plugin discovery couldn't be skipped for it. Added.
6. tests/hermes_cli/test_tools_config.py — my fix(tools): run _run_post_setup() in _reconfigure_provider() for parity (salvage #26642) #27306 (post_setup parity) parametrized over four browser providers, but the catalog was refactored to keep only Camofox + the consolidated Nous Subscription/Local Browser pair (the latter two have no env_vars so don't exercise this code path). Pruned the stale parametrize entries to just Camofox.

Validation

• scripts/run_tests.sh tests/gateway/test_google_chat.py tests/gateway/test_teams.py tests/hermes_cli/test_tools_config.py tests/hermes_cli/test_startup_plugin_gating.py tests/gateway/test_background_command.py tests/gateway/test_telegram_thread_fallback.py tests/gateway/test_voice_command.py tests/acp/test_registry_manifest.py tests/run_agent/test_provider_parity.py -q → 646/646 pass.
• ruff check clean.

Remaining main failures (NOT fixed by this PR)

Three failures remain that are unrelated to the salvage train and look like genuine pre-existing production regressions:

• tests/hermes_cli/test_model_switch_custom_providers.py::test_list_groups_same_name_custom_providers_into_one_row — list_authenticated_providers is leaking the global model catalog into custom-provider rows when grouping by name. Real bug.
• tests/tools/test_voice_cli_integration.py::TestVprintForceParameter::test_error_messages_use_force_in_run_agent — expects at least one _vprint(..., force=True) for ❌-prefixed error messages in run_agent.py; found zero. Someone added an error path without force=True.
• tests/run_agent/test_deepseek_reasoning_content_echo.py::TestNeedsKimiToolReasoning::test_non_kimi_provider — deepseek reasoning-content echo guard regressed.

Each deserves its own targeted PR. Filing follow-ups separately.

[github-actions]

🔎 Lint report: hermes/salvage-tier2 vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 8708 on HEAD, 8708 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 4590 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.