Problem

A provider can crash (SIGSEGV) while claiming rewards:

RewardServer.sendRewardsClaim (goroutine)
  └─ ProviderTxSender.TxRelayPayment
       └─ SimulateAndBroadCastTxWithRetryOnSeqMismatch
            └─ simulateTxWithRetry
                 └─ tx.CalculateGas   💥 nil pointer dereference

cosmos-sdk's CalculateGas does simRes.GasInfo.GasUsed with no nil check (client/tx/tx.go). A node can return a successful Simulate response with GasInfo == nil (the proto field is optional), so the dereference panics — addr=0x8 in the signal is exactly the offset of GasUsed inside GasInfo. The reward claim runs in an un-recovered goroutine, so that panic crashes the entire lavap provider process (downtime, missed relays, lost claim).

Fix (lavap-side, two layers)

1. simulateTxWithRetry now calls calculateGasWithRecover, which wraps tx.CalculateGas and converts a panic into a normal error handled by the existing retry/fail path. It lives in the shared simulation path, so it protects every gas-estimated tx (consumer + provider), not just reward claims.
2. A backstop recover() in the reward-claim goroutine logs the panic + stack and marks the claim for retry, so no panic in that async path can take the provider down.

Normal behavior is unchanged.

Verification

• New regression test TestSendRewardsClaim_RecoversFromTxPanic (panicking TxRelayPayment mock): proven to fail without the change (the goroutine panic crashes the test binary) and pass with it; the recovered claim is marked for retry.
• Full rewardserver + statetracker packages, go vet, gofmt all clean.

Related (not in this PR)

The root nil-deref is in the cosmos-sdk fork (github.com/lavanet/cosmos-sdk, client/tx/tx.go): CalculateGas should nil-check simRes.GasInfo and return an error instead of dereferencing. That's the complementary deeper fix in the fork; this PR is the lavap-side guard so a bad node response degrades gracefully regardless. Independent of #2301/#2302 (different subsystem).

🤖 Generated with Claude Code