feat(eval): add rerun-failed runner pooling

apps/cli/src/commands/eval/commands/run.ts

@@ -168,10 +168,11 @@ export const evalRunCommand = command({
       description:
         'Resume an interrupted run: skip already-completed tests and append new results to --output dir',
     }),
-    rerunFailed: flag({
+    rerunFailed: option({
+      type: optional(string),
       long: 'rerun-failed',
       description:
-        'Rerun failed/errored tests while keeping passing results. Implies --resume semantics',
+        'Run ID, run workspace, or index.jsonl to rerun failed/errored tests while keeping passing results',
     }),
     strict: flag({
       long: 'strict',

apps/cli/src/commands/eval/run-cache.ts

@@ -54,10 +54,10 @@ export async function loadRunCache(cwd: string): Promise<RunCache | undefined> {
 /**
  * Resolve the cached last-run directory for a cwd, if it still exists on disk.
  * Returns undefined when there is no cache, the cache lacks a `lastRunDir`,
- * or the directory has since been deleted. Used by `--resume` / `--rerun-failed`
- * to default `--output` to the most recent run when no explicit dir is given,
- * matching the convention used by promptfoo (`--resume [evalId]`) and
- * OpenCompass (`-r [timestamp]`).
+ * or the directory has since been deleted. Used by `--resume` to default
+ * `--output` to the most recent run when no explicit dir is given, matching
+ * the convention used by promptfoo (`--resume [evalId]`) and OpenCompass
+ * (`-r [timestamp]`).
  */
 export async function resolveCachedRunDir(cwd: string): Promise<string | undefined> {
   const cache = await loadRunCache(cwd);