fix(stages): only iterate over active stages

[joshieDo]

During engine sync, we iterate over all stages in StageId::ALL and bump their checkpoints via update_pipeline_stages. During the pipeline consistency check, we iterate all checkpoints and compare them to the header stage.

However, the pipeline itself conditionally adds PruneSenderRecoveryStage only if sender pruning is enabled:

reth/crates/stages/stages/src/sets.rs

Lines 337 to 340 in b79c58d

	// If sender recovery prune mode is set, add the prune sender recovery stage.
	.add_stage_opt(self.prune_modes.sender_recovery.map(|prune_mode| {
	PruneSenderRecoveryStage::new(prune_mode, self.stages_config.prune.commit_threshold)
	}))

This can result in the following weird scenario if sender pruning is not enabled:

1. PruneSenderRecovery checkpoint is set to block 100 by engine sync (via update_pipeline_stages)
2. Run unwind command to block 95. Since PruneSenderRecoveryStage is not in the pipeline, its checkpoint stays at 100 (ahead of other stages, but not an issue yet)
3. Run pipeline forward to block 105. Once again, PruneSenderRecoveryStage is not in the pipeline and won't run, so its checkpoint stays at 100
4. Now PruneSenderRecovery (100) < Headers (105), which triggers the inconsistency check:

or

1. Engine sync to block 50 → update_pipeline_stages creates PruneSenderRecovery at 50
2. Pipeline sync continues to block 100 → Headers at 100, but PruneSenderRecovery stuck at 50
3. CLI unwind to block 60 → Headers at 60, PruneSenderRecovery still at 50
4. reth node → consistency check: 50 < 60

reth/crates/node/builder/src/launch/common.rs

Lines 975 to 985 in b79c58d

	if stage_checkpoint < first_stage_checkpoint {
	debug!(
	target: "consensus::engine",
	first_stage_id = %first_stage,
	first_stage_checkpoint,
	inconsistent_stage_id = %stage_id,
	inconsistent_stage_checkpoint = stage_checkpoint,
	"Pipeline sync progress is inconsistent"
	);
	return self.blockchain_db().block_hash(first_stage_checkpoint);
	}

[mediocregopher]

Makes sense to me 👍

[joshieDo]

was not sure how to bring era_enabled here

hardcoding it to truth here, results in the same behaviour as before. And should be fine since we filter it out anyway in the pipeline consistency check.