Optionally handle "repo not found" megarepo errors by panicing

Summary:
## This stack

S460221 followups to fix bugs around sharding as well as general brittleness.

## This diff

We've had multiple SEVs where we reported "repo not found 2135" to grepo sync after releasing changes. Panicking helps by causing Conveyor to revert immediately. These errors are retried by the client, meaning a rollback would stay below the SEV threshold.

Reviewed By: mzr

Differential Revision: D64702687

fbshipit-source-id: 247fd12f9c8cdaea1c6737c35248d8ffadfb2440

eden/mononoke/megarepo_api/src/lib.rs

@@ -242,7 +242,20 @@ impl<R: MononokeRepo> MegarepoApi<R> {
             .repo_by_id(ctx.clone(), repo_id)
             .await
             .map_err(MegarepoError::internal)?
-            .ok_or_else(|| MegarepoError::request(anyhow!("repo not found {}", repo_id)))?
+            .ok_or_else(|| {
+                if repo_id.id() != 0 // special case for some tests
+                    && justknobs::eval(
+                        "scm/mononoke:megarepo_panic_on_repo_not_found_error",
+                        None,
+                        Some(&repo_id.to_string()),
+                    )
+                    .unwrap_or(false)
+                {
+                    panic!("repo not found {}", repo_id)
+                } else {
+                    MegarepoError::request(anyhow!("repo not found {}", repo_id))
+                }
+            })?
             .with_authorization_context(AuthorizationContext::new_bypass_access_control())
             .build()
             .await?;