perf(stages): avoid v2 history count pre-pass

[yongkangc]

storage_v2 history indexing was paying a near-fixed per-cycle cost even when only a small live tail needed processing. Both history stages called account_changeset_count() / storage_changeset_count() only to format progress logs, and those helpers scanned .csoff metadata across finalized static files.

This removes that pre-pass from the live-sync path by logging progress against the requested block range instead. It also makes the count helpers themselves cheaper by reading only the last committed .csoff record per static file, so remaining callers no longer materialize every offset record.

Repro gist: https://gist.github.com/yongkangc/06d1da54a32b4a1818c9432483fef7a6

Repro setup:

• cargo test -p reth-stages repro_collect_bottleneck_breakdown -- --ignored --nocapture
• 200k historical blocks
• 1k blocks per static file
• 36-block live tail

Metric	Before	After stage fix	After combined fix
Account stage	35.2ms	1.5ms	1.6ms
Storage stage	35.6ms	2.7ms	2.4ms
Account count probe	33.4ms	32.9ms	2.7ms
Storage count probe	35.8ms	32.7ms	2.6ms

Co-Authored-By: YK 46377366+yongkangc@users.noreply.github.com

Prompted by: yongkangc

[github-actions]

⚠️ Changelog not found.

A changelog entry is required before merging. We've generated a suggested changelog based on your changes:

Preview

---
reth-stages: patch
reth-provider: patch
---

Fixed changeset count calculation to use last offset entry instead of summing all offsets, and improved history index progress reporting to track blocks in the active range rather than total changesets. Added tests for account and storage changeset count correctness.

Add changelog to commit this to your branch.

[yongkangc]

Closing this in favor of #23296, which addresses the same bottleneck with a smaller change. It makes O(files) instead of O(blocks), preserves the existing progress accounting, and fixes the hot path without changing the stage collection logic.

[yongkangc]

Closing this in favor of #23296, which addresses the same bottleneck with a smaller change. It makes changeset_count() O(files) instead of O(blocks), preserves the existing progress accounting, and fixes the hot path without changing the stage collection logic.