Parent issue

Part of #16239 -- avoiding inconsistency due to lexicographic prune.

Problem

DomainLatestIterFile.initCursorMDBX() and advanceInFiles() in db/state/domain_stream.go push DB entries into the priority heap without checking whether their step falls within the file range. The comment // DB can store not-finished step, it means - then set first txn in step - it anyway will be ahead of files is incorrect for prunable data — DB entries whose step overlaps with files have step * stepSize < files.EndTxNum().

Affected locations

• initCursorMDBX() LargeValues path (line ~210)
• initCursorMDBX() DupSort path (line ~228)
• advanceInFiles() LargeValues DB_CURSOR branch (line ~292)
• advanceInFiles() DupSort DB_CURSOR branch (line ~312)

All four locations have the same misleading comment and lack a filesEndTxNum check.

Fix

Apply the same "skip DB entries within file range" pattern used in:

• debugIteratePrefixLatest: skip DB entries within file range #20355 (debugIteratePrefixLatest)
• db/state: RangeAsOf DB iterator starts from files.EndTxNum() #20360 (RangeAsOf)
• iterateChangedRecent: skip DB entries within file range #20361 (iterateChangedRecent)
• recentIterateRange: narrow DB query to exclude file range #20362 (recentIterateRange)

DomainLatestIterFile would need a filesEndTxNum uint64 field set during init() from domainRoTx.files.EndTxNum(), since advanceInFiles() doesn't have access to domainRoTx.