fix: prevent model selector label degradation after first conversation

After the first conversation, captureCapabilities() caches the SDK's
supportedModels() which have different displayNames ("Sonnet", "Opus 4")
than our catalog ("Sonnet 4.6", "Opus 4.6"). The /api/providers/models
route was using m.displayName directly from the SDK cache, overriding
the correct labels.

Fix: prefer CLAUDE_MODELS[alias].displayName when available, fall back
to SDK's displayName for unknown/future models.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: jasonjaclyn2017

src/app/api/providers/models/route.ts

@@ -1,6 +1,7 @@
 import { NextResponse } from 'next/server';
 import { getAllProviders, getDefaultProviderId, setDefaultProviderId, getProvider, getModelsForProvider, getSetting } from '@/lib/db';
 import { getContextWindow } from '@/lib/model-context';
+import { CLAUDE_MODELS } from '@/lib/model-ids';
 import { getDefaultModelsForProvider, inferProtocolFromLegacy, findPresetForLegacy } from '@/lib/provider-catalog';
 import type { Protocol } from '@/lib/provider-catalog';
 import type { ErrorResponse, ProviderModelGroup } from '@/types';
@@ -73,7 +74,7 @@ export async function GET() {
           const cw = getContextWindow(m.value);
           return {
             value: m.value,
-            label: m.displayName,
+            label: CLAUDE_MODELS[m.value as keyof typeof CLAUDE_MODELS]?.displayName ?? m.displayName,
             description: m.description,
             supportsEffort: m.supportsEffort,
             supportedEffortLevels: m.supportedEffortLevels,