Add backoff retry mechanism for head tipset synchronization between Forest and Lotus

[coderabbitai]

Problem

During integration testing between Forest and Lotus nodes, test failures can occur when the nodes have different head tipsets. This manifests as discrepancies where Forest returns not indexed while Lotus returns valid for the same request.

Example Test Failure

From CI run: https://github.com/ChainSafe/forest/actions/runs/21770981495/job/62818291671#step:5:1199

{
  "request_params": [
    "0x8e037f5e7fb7e5b8516c405527c2e9b60f8ef5dedcc30d719e37ae9fc1db393f",
    800
  ],
  "forest_status": {
    "rejected": "not indexed"
  },
  "lotus_status": "valid",
  "execution_duration_ms": 30
}

Proposed Solution

Implement a backoff retry mechanism to mitigate such failures when head tipsets are temporarily out of sync. This would allow tests to retry requests with exponential backoff, giving nodes time to synchronize.

Context

• Related discussion: [automated] RPC parity test failure @ 7/2/26 00:52 #6558
• Requested by: @hanabi1224