[FEATURE] Add testapp throughput benchmark to isolate ev-node overhead

[chatton]

Problem to solve

No dedicated benchmark isolating ev-node's sequencer loop overhead independent of the EVM executor.

Proposed solution or feature

Add an E2E throughput benchmark using ev-node + testapp (KV executor). Since the KV executor is trivially fast (key-value upserts, no gas model), any measured latency is attributable to ev-node itself (sequencing, DA submission, block management).

• use a Go-native load generator (HTTP POST key=value to testapp) since spamoor requires an EVM
• measure blocks/s, TPS, block intervals with tracing enabled
• capture latency breakdown: ProduceBlock, ExecuteTxs (basically nothing), DA.Submit, GetNextBatch

Alternatives considered

The ev-reth EVM benchmarks already compute ev-node overhead via ProduceBlock - ExecuteTxs from traces. This testapp benchmark would validate that number independently but may have minimal additional value if the trace-derived overhead is consistent. Consider deprioritizing unless overhead numbers from ev-reth benchmarks look suspicious.

Additional context

Part of #2288