API Rate Limit Not Properly Handled - Model Stops Responding After Few Queries #11104

 **Environment:**
 - API Provider: https://build.nvidia.com/
 - Rate Limit: 40 rpm
 - Models Affected: GLM-4.7 or other models from NVIDIA API catalog                                                                                                                             

 **Steps to Reproduce:**
 1. Start a conversation with the model (e.g., GLM-4.7 or other models)
 2. Ask 2-3 questions successfully                                                                                                                                                                
 3. On subsequent questions, the model stops responding                                                                                                                                           
 4. Animation continues running indefinitely                                                                                                                                                      
 5. All submitted tasks remain stuck in the task list
                                                                                                                                                                                                  
 **Expected Behavior:**
 - Model should respond to all queries within rate limits
 - If rate limit is exceeded, user should receive clear error message
 - System should handle rate limit gracefully

 **Actual Behavior:**
 - Model silently fails after 2-3 queries                                                                                                                                                         
 - Animation continues but no response is ever returned                                                                                                                                           
 - Issue persists for several hours                                                                                                                                                               
 - No error message displayed to indicate rate limit was reached
                                                                                                                                                                                                  
 **Analysis:**                                                                                                                                                                                    
 It appears the API returns HTTP 429 (Too Many Requests) or similar rate limit error when usage exceeds 40 rpm, but the client (opencode) does not properly detect or display this error,         
 leaving the user in a confused state where the UI shows activity but no actual response is generated.

 **Request:**
 Please implement proper rate limit error handling that:
 1. Detects rate limit responses from the API
 2. Displays a clear error message to the user                                                                                                                                                  
 3. Shows remaining quota or wait time   
 4. Prevents indefinite hanging states   
 5. Provides proper timeout behavior instead of leaving tasks stuck