fix(model-switch): normalize Unicode dashes from Telegram/iOS input

Telegram on iOS auto-converts double hyphens (--) to em dashes (—)
or en dashes (–) via autocorrect. This breaks /model flag parsing
since parse_model_flags() only recognizes literal '--provider' and
'--global'.

When the flag isn't parsed, the entire string (e.g. 'glm-5.1 —provider zai')
gets treated as the model name and fails with 'Model names cannot
contain spaces.'

Fix: normalize Unicode dashes (U+2012-U+2015) to '--' when they
appear before flag keywords (provider, global), before flag extraction.

The existing test suite in test_model_switch_provider_routing.py
already covers all four dash variants — this commit adds the code
that makes them pass.

hermes_cli/model_switch.py

@@ -274,6 +274,11 @@ def parse_model_flags(raw_args: str) -> tuple[str, str, bool]:
     is_global = False
     explicit_provider = ""
 
+    # Normalize Unicode dashes (Telegram/iOS auto-converts -- to em/en dash)
+    # A single Unicode dash before a flag keyword becomes "--"
+    import re as _re
+    raw_args = _re.sub(r'[\u2012\u2013\u2014\u2015](provider|global)', r'--\1', raw_args)
+
     # Extract --global
     if "--global" in raw_args:
         is_global = True