Add DeepSeek V4 flash/pro support and DeepSeek thinking compatibility (#877)

* Add DeepSeek V4 support and thinking compatibility

* Fix DeepSeek profile persistence regression

* Align multi-model handling with openai-multi-model

src/utils/providerProfile.ts

@@ -15,6 +15,7 @@ import {
 } from './providerRecommendation.js'
 import { readGeminiAccessToken } from './geminiCredentials.js'
 import { getOllamaChatBaseUrl } from './providerDiscovery.js'
+import { getPrimaryModel } from './providerModels.js'
 import { getProviderValidationError } from './providerValidation.js'
 import {
   maskSecretForDisplay,