chain(ethereum): treat Reth StackUnderflow/StackOverflow/OpcodeNotFound halts as deterministic

[cargopete]

Fixes #6582.

Problem

Reth surfaces EVM halts through reth's EthApiError::EvmHalt(HaltReason), which is formatted with the reason's Debug representation:

// reth: crates/rpc/rpc-eth-types/src/error/mod.rs
#[error("EVM error: {0:?}")]
EvmHalt(HaltReason),

revm's HaltReason / InstructionResult variants are derived Debug, so they render in CamelCase with no spaces. A stack underflow therefore reaches eth_call error handling as:

EVM error: StackUnderflow

is_rpc_revert_message lower-cases the message and checks contains, so it compares against "evm error: stackunderflow". The existing space-separated entry "stack underflow" never matches this, so the error is treated as non-deterministic — the subgraph stalls/retries instead of recording a deterministic revert.

This is the same gap streamingfast/eth-go#10 closed on the Firehose side.

Fix

Add the missing reth halt variants (no spaces, matching the Debug output) to RPC_EXECUTION_ERRORS:

• stackunderflow ← the reported bug
• stackoverflow
• opcodenotfound

The list already contains invalidjump and invalidfeopcode, which catch "EVM error: InvalidJump" / "InvalidFEOpcode" via this exact path — so the pattern is already established and proven; these three siblings were simply missed.

Note: reth's OutOfGas is intercepted before EvmHalt and rendered as "out of gas: ..." (with spaces), which the existing "out of gas" entry already matches — so unlike the Firehose list, a no-space outofgas is not required here.

Verification

• reth: EvmHalt uses {0:?} (crates/rpc/rpc-eth-types/src/error/mod.rs).
• revm: InstructionResult / HaltReason are #[derive(Debug)] with variants StackUnderflow, StackOverflow, OpcodeNotFound, InvalidJump, InvalidFEOpcode (crates/interpreter/src/instruction_result.rs).
• Precedent: Categorize StackUnderflow Reth error as deterministic streamingfast/eth-go#10.

[incrypto32]

Thanks for the PR @cargopete.
Branch protection here requires signed commits, so the merge’s blocked. Mind re-signing and force-pushing this commit?

[cargopete]

Thanks for the PR @cargopete.

Branch protection here requires signed commits, so the merge’s blocked. Mind re-signing and force-pushing this commit?

Sure thing, will do later today

[cargopete]

@incrypto32 should be done now