fix(providers): capture claude-cli stream errors and surface real message (#884)

claude-cli emits API errors (rate limits, usage limits) via stream-json
stdout result events, not stderr. Previous code only checked
ev.Subtype == "error" and used an empty stderr buffer in the final
wrap, so users saw a generic "something went wrong" instead of the
real cause.

- Add IsError and Error fields to cliStreamEvent to parse both signals
- Capture stream error message when FinishReason flips to "error"
- Substitute captured message when stderr is empty in cmd.Wait path
- Extend gateway rate-limit classifier to recognize "usage limit"

cmd/gateway_errors.go

@@ -32,7 +32,7 @@ func formatAgentError(err error) string {
 	}
 
 	// 4. Rate limit
-	if containsAny(lower, "rate limit", "rate_limit", "too many requests", "429", "quota exceeded", "resource_exhausted") {
+	if containsAny(lower, "rate limit", "rate_limit", "too many requests", "429", "quota exceeded", "resource_exhausted", "usage limit") {
 		return "⚠️ API rate limit reached. Please try again later."
 	}
 

internal/providers/claude_cli_types.go

@@ -29,6 +29,8 @@ type cliStreamEvent struct {
 	Model   string        `json:"model,omitempty"`
 	CostUSD float64       `json:"cost_usd,omitempty"`
 	Usage   *cliUsage     `json:"usage,omitempty"`
+	IsError bool          `json:"is_error,omitempty"` // true when subtype="error"
+	Error   string        `json:"error,omitempty"`    // error message (may be set when result is empty)
 }
 
 // cliStreamMsg wraps content blocks inside an assistant message event.