fix(/model): show provider-enforced context length, not raw models.dev (#15438)

/model gpt-5.5 on openai-codex showed 'Context: 1,050,000 tokens' because
the display block used ModelInfo.context_window directly from models.dev.
Codex OAuth actually enforces 272K for the same slug, and the agent's
compressor already runs at 272K via get_model_context_length() — so the
banner + real context budget said 272K while /model lied with 1M.

Route the display context through a new resolve_display_context_length()
helper that always prefers agent.model_metadata.get_model_context_length
(which knows about Codex OAuth, Copilot, Nous caps) and only falls back
to models.dev when that returns nothing.

Fix applied to all 3 /model display sites:
  cli.py _handle_model_switch
  gateway/run.py picker on_model_selected callback
  gateway/run.py text-fallback confirmation

Reported by @emilstridell (Telegram, April 2026).

tests/hermes_cli/test_model_switch_context_display.py

@@ -0,0 +1,90 @@
+"""Regression test for /model context-length display on provider-capped models.
+
+Bug (April 2026): `/model gpt-5.5` on openai-codex (ChatGPT OAuth) showed
+"Context: 1,050,000 tokens" because the display code used the raw models.dev
+``ModelInfo.context_window`` (which reports the direct-OpenAI API value) instead
+of the provider-aware resolver. The agent was actually running at 272K — Codex
+OAuth's enforced cap — so the display was lying to the user.
+
+Fix: ``resolve_display_context_length()`` prefers
+``agent.model_metadata.get_model_context_length`` (which knows about Codex OAuth,
+Copilot, Nous, etc.) and falls back to models.dev only if that returns nothing.
+"""
+from __future__ import annotations
+
+from unittest.mock import patch
+
+from hermes_cli.model_switch import resolve_display_context_length
+
+
+class _FakeModelInfo:
+    def __init__(self, ctx):
+        self.context_window = ctx
+
+
+class TestResolveDisplayContextLength:
+    def test_codex_oauth_overrides_models_dev(self):
+        """gpt-5.5 on openai-codex must show Codex's 272K cap, not models.dev's 1.05M."""
+        fake_mi = _FakeModelInfo(1_050_000)  # what models.dev reports
+        with patch(
+            "agent.model_metadata.get_model_context_length",
+            return_value=272_000,  # what Codex OAuth actually enforces
+        ):
+            ctx = resolve_display_context_length(
+                "gpt-5.5",
+                "openai-codex",
+                base_url="https://chatgpt.com/backend-api/codex",
+                api_key="",
+                model_info=fake_mi,
+            )
+        assert ctx == 272_000, (
+            "Codex OAuth's 272K cap must win over models.dev's 1.05M for gpt-5.5"
+        )
+
+    def test_falls_back_to_model_info_when_resolver_returns_none(self):
+        fake_mi = _FakeModelInfo(1_048_576)
+        with patch(
+            "agent.model_metadata.get_model_context_length", return_value=None
+        ):
+            ctx = resolve_display_context_length(
+                "some-model",
+                "some-provider",
+                model_info=fake_mi,
+            )
+        assert ctx == 1_048_576
+
+    def test_returns_none_when_both_sources_empty(self):
+        with patch(
+            "agent.model_metadata.get_model_context_length", return_value=None
+        ):
+            ctx = resolve_display_context_length(
+                "unknown-model",
+                "unknown-provider",
+                model_info=None,
+            )
+        assert ctx is None
+
+    def test_resolver_exception_falls_back_to_model_info(self):
+        fake_mi = _FakeModelInfo(200_000)
+        with patch(
+            "agent.model_metadata.get_model_context_length",
+            side_effect=RuntimeError("network down"),
+        ):
+            ctx = resolve_display_context_length(
+                "x", "y", model_info=fake_mi
+            )
+        assert ctx == 200_000
+
+    def test_prefers_resolver_even_when_model_info_has_larger_value(self):
+        """Invariant: provider-aware resolver is authoritative, even if models.dev
+        reports a bigger window."""
+        fake_mi = _FakeModelInfo(2_000_000)
+        with patch(
+            "agent.model_metadata.get_model_context_length", return_value=128_000
+        ):
+            ctx = resolve_display_context_length(
+                "capped-model",
+                "capped-provider",
+                model_info=fake_mi,
+            )
+        assert ctx == 128_000