fix: delete consumed pendingFailureDurations entries to prevent stale leaks

The pendingFailureDurations map was never cleaned up after a duration
was consumed, allowing stale entries to attach incorrect durations to
later failures with the same suite/test key in retry scenarios.

Also adds a clarifying comment on the Swift Testing issue-count-as-
failed-count approximation.

Author: cameroncooke

src/utils/swift-testing-line-parsers.ts

@@ -110,6 +110,11 @@ export function parseSwiftTestingRunSummary(line: string): ParsedTotals | null {
   const total = Number(match[1]);
   const durationText = `${match[2]}s`;
 
+  // Swift Testing reports "issues" not "failed tests" -- a single test can produce
+  // multiple issues (e.g. multiple #expect failures). This is the best available
+  // approximation; the framework doesn't report a distinct failed-test count in its
+  // summary line. Downstream reconciliation via Math.max(failedTests, testFailures.length)
+  // partially mitigates overcounting.
   const issueMatch = line.match(/with (\d+) issues?/u);
   const failed = issueMatch ? Number(issueMatch[1]) : 0;
 

src/utils/xcodebuild-event-parser.ts

@@ -176,6 +176,7 @@ export function createXcodebuildEventParser(options: EventParserOptions): Xcodeb
 
     const durationMs = pendingFailureDurations.get(key);
     if (durationMs !== undefined) {
+      pendingFailureDurations.delete(key);
       emitFailureEvent({ ...failure, durationMs });
       return;
     }
@@ -211,6 +212,7 @@ export function createXcodebuildEventParser(options: EventParserOptions): Xcodeb
       emitFailureEvent({ ...failure, durationMs });
     }
     pendingFailureDiagnostics.delete(key);
+    pendingFailureDurations.delete(key);
   }
 
   function emitTestProgress(): void {