fix: widen base-URL hostname identity class to remaining substring sites

Follow-up to #85737, which migrated five provider-identity sites onto
utils.base_url_host_matches()/base_url_hostname(). This completes the class
sweep (never-patch-predicates: one owner, every site) and folds in the two
open contributor PRs attacking individual sites:

- agent/auxiliary_client.py ZAI/Kimi OpenAI-wire rewrite (PR #85715,
  pierrenode): 'bigmodel'/'api.z.ai'/'api.kimi.com' substring checks
  rewrote proxy paths containing those markers.
- hermes_cli/runtime_provider.py Azure endpoint detection (PR #74721,
  RelaxJonh, issue #74312): 'azure.com' substring picked the Azure key
  for non-Azure hosts whose path contained the text.
- run_agent.py: _is_azure_openai_url, _is_copilot_url, Anthropic
  credential-refresh azure guard, _anthropic_preserve_dots host
  allowlist, OpenRouter/mistral reasoning gates.
- agent/chat_completion_helpers.py: nousresearch / nvidia detection.
- agent/conversation_loop.py: GitHub Models 413 hint.
- agent/usage_pricing.py: localhost billing-route detection.
- hermes_cli/model_switch.py: api.openai.com catalog fallback and
  localhost custom-provider detection.
- cli.py: local-model autodetect and Ollama/LM Studio context-length
  hints (port-anchored instead of '11434' in URL).
- tools/mcp_oauth.py: Figma remote-MCP detection.
- tools/skills_hub.py: raw.githubusercontent.com source-URL check.

Regression tests extend tests/hermes_cli/test_base_url_host_identity.py
(azure/copilot/dotted-model/figma proxy-path + lookalike cases) and
tests/agent/test_minimax_auxiliary_url.py (ZAI/Kimi path false positives).

Closes #74312. Salvages #85715 and #74721 with authorship preserved.

cli.py

@@ -224,7 +224,7 @@ from hermes_cli.browser_connect import (
     try_launch_chrome_debug,
 )
 from hermes_cli.env_loader import load_hermes_dotenv
-from utils import base_url_host_matches, fast_safe_load
+from utils import base_url_host_matches, base_url_hostname, fast_safe_load
 
 _hermes_home = get_hermes_home()
 _project_env = Path(__file__).parent / '.env'
@@ -4604,7 +4604,7 @@ class HermesCLI(CLIAgentSetupMixin, CLICommandsMixin, CLIBillingMixin):
         # Auto-detect model from local server if still on default
         if self.model == _DEFAULT_CONFIG_MODEL:
             _base_url = (_model_config.get("base_url") or "") if isinstance(_model_config, dict) else ""
-            if "localhost" in _base_url or "127.0.0.1" in _base_url:
+            if base_url_hostname(_base_url) in ("localhost", "127.0.0.1"):
                 from hermes_cli.runtime_provider import _auto_detect_local_model
                 _detected = _auto_detect_local_model(_base_url)
                 if _detected:
@@ -7834,11 +7834,18 @@ class HermesCLI(CLIAgentSetupMixin, CLICommandsMixin, CLIBillingMixin):
                 f"[dim]   Hermes needs at least {MINIMUM_CONTEXT_LENGTH:,} tokens. Tool schemas + system prompt use a large fixed prefix.[/]"
             )
             base_url = getattr(self, "base_url", "") or ""
-            if "11434" in base_url or "ollama" in base_url.lower():
+            from urllib.parse import urlparse as _urlparse
+            try:
+                _parsed = _urlparse(base_url if "://" in base_url else f"//{base_url}")
+                _port = _parsed.port
+            except ValueError:
+                _port = None
+            _host = base_url_hostname(base_url)
+            if _port == 11434 or "ollama" in _host:
                 self._console_print(
                     f"[dim]   Ollama fix: OLLAMA_CONTEXT_LENGTH={MINIMUM_CONTEXT_LENGTH} ollama serve[/]"
                 )
-            elif "1234" in base_url:
+            elif _port == 1234:
                 self._console_print(
                     "[dim]   LM Studio fix: Set context length in model settings → reload model[/]"
                 )