fix: move JournalInspectTr to host.rs with proper error stashing

[Troublor]

Summary

• Bug fix: Custom instruction wrappers in instructions.rs previously had a bug where some paths halted with FatalExternalError without stashing the underlying DB error in ctx.error(), causing panics when revm tried to drain the stashed error.
• Refactoring: Moved JournalInspectTr trait to host.rs with two implementations:
  • Journal<DB>: returns Result<T, DB::Error> for direct error propagation
  • MegaContext: stashes errors in self.error() as ContextError::Custom and returns Err(())
• Simplified all 4 instruction call sites from manual error stashing to let Ok(...) = context.host.inspect_*(...) else { halt(FatalExternalError); return; }
• Changed HostExt gas methods (sstore_set_storage_gas, new_account_storage_gas, create_contract_storage_gas) to return Option instead of Result, stashing errors internally (consistent with revm's Host trait pattern)
• Replaced load_account with inspect_account_delegated in storage_gas_ext::create to avoid marking the creator account warm
• Added ErrorInjectingDatabase test utility for simulating DB errors
• Added 3 integration tests verifying that DB errors during inspect_storage (SSTORE) and inspect_account_delegated (CALL, STATICCALL) are properly stashed and surfaced as EVMError::Custom

Test plan

• cargo check -p mega-evm passes
• cargo check -p mega-evm --target riscv64imac-unknown-none-elf --no-default-features passes (no_std)
• cargo test -p mega-evm — all tests pass (including 3 new db_error tests)
• cargo clippy --workspace --lib --examples --tests --benches --all-features --locked — no warnings
• cargo fmt --all --check — formatted