fix(provider): mark DeepSeek models as BYOK (#162)

Author: Vizards

src/provider/models.ts

@@ -7,7 +7,8 @@ import { toModelCostInfo, type ModelCostInformation } from './pricing/costs';
  * NOTE: Non-public API surface.
  *
  * The fields below (`configurationSchema` on chat info, cost metadata,
- * `modelConfiguration` on response options, plus `isUserSelectable` / `statusIcon`)
+ * `modelConfiguration` on response options, plus `isBYOK` / `isUserSelectable` /
+ * `statusIcon`)
  * are not part of the stable `vscode.LanguageModelChat*` typings yet. They are
  * the same shape currently consumed by GitHub Copilot Chat to render model picker
  * metadata and per-model configuration controls.
@@ -25,6 +26,7 @@ type ThinkingEffortConfigurationSchema = ReturnType<typeof buildThinkingEffortSc
 export type ModelPickerChatInformation = vscode.LanguageModelChatInformation &
 	ModelCostInformation & {
 		readonly isUserSelectable: boolean;
+		readonly isBYOK: true;
 		readonly statusIcon?: vscode.ThemeIcon;
 		readonly configurationSchema?: ThinkingEffortConfigurationSchema;
 	};
@@ -46,6 +48,7 @@ export function toChatInfo(
 		statusIcon: hasApiKey ? undefined : new vscode.ThemeIcon('warning'),
 		maxInputTokens: m.maxInputTokens,
 		maxOutputTokens: m.maxOutputTokens,
+		isBYOK: true,
 		isUserSelectable: true,
 		capabilities: {
 			toolCalling: m.capabilities.toolCalling,