Handle ZCONNECTIONLOSS with exponential backoff

Summary:
Zeus ZCONNECTIONLOSS errors are used for loadshedding (see
https://fb.workplace.com/groups/zeus.users/permalink/31336632699291946/).
Currently these fall through to `InternalError::Other`, which uses quadratic
backoff starting at 100ms — the worker crashes after 5 consecutive failures
in ~3 seconds. This is too aggressive and doesn't give Zeus enough breathing
room to recover.

This diff adds a dedicated `TransientZeusError` variant to `InternalError`
that matches `ZCONNECTIONLOSS` and applies exponential backoff starting at
500ms (500ms, 1s, 2s, 4s, then crash). This extends the total retry window
from ~3s to ~7.5s before the worker crashes.

This is a mitigation while we request more Zelos capacity.

Alert: https://fburl.com/onedetection/yrvpdgcc

Reviewed By: YousefSalama

Differential Revision: D99992330

fbshipit-source-id: 28193c9f6744579b89d38e38c59c961c9050513e

Author: Clara Rull

eden/mononoke/repo_attributes/repo_derivation_queues/src/errors.rs

@@ -31,6 +31,8 @@ pub enum InternalError {
     ItemDeleted(String),
     #[error("Attepmt to create Derivation Item with dependency on itself {0:#?}")]
     CircularDependency(DagItemId),
+    #[error("Transient Zeus connection error: {0}")]
+    TransientZeusError(String),
     #[error(transparent)]
     Other(#[from] anyhow::Error),
 }
@@ -55,6 +57,10 @@ impl From<zeus_client::ZeusError> for InternalError {
                 message: msg,
                 exception_type: ZelosExceptionType::ZNONODE,
             } => InternalError::ItemDeleted(msg),
+            zeus_client::ZeusError::RuntimeError {
+                message: msg,
+                exception_type: ZelosExceptionType::ZCONNECTIONLOSS,
+            } => InternalError::TransientZeusError(msg),
             _ => InternalError::Other(e.into()),
         }
     }