HistoryRange: perf

[AskAlexSharov]

---

1. heap.Fix instead of heap.Pop + heap.Push

The k-way merge uses a min-heap to track the current position in each segment file. The original code for each key:
top := heap.Pop(&hi.h) // remove root, sift-down: O(log n)
// ... advance reader ...
heap.Push(&hi.h, top) // re-insert, sift-up: O(log n)
We can instead just update the root in-place:
top := hi.h[0] // peek — no removal
// ... advance reader ...
heap.Fix(&hi.h, 0) // only sift-down: O(log n)
Fix is cheaper because it skips the sift-up step that Push always does. Works because the updated key is always ≥ the old one (files
are sorted ascending).
(-10%)

---

2. File index cached in ReconItem

For every key that passes the filter, the original code searched for the corresponding history file:
historyItem, ok := hi.hc.getFileDeprecated(top.startTxNum, top.endTxNum)
// linear scan: for i := range ht.files { if files[i].startTxNum == ... }
With 60 files, that's 60 comparisons per key. Since each ReconItem already knows its file range (startTxNum/endTxNum), we look up the
index once at construction time and store it:
heap.Push(&h, &ReconItem{..., histFileIdx: j}) // done once per file
// ...
historyItem := hi.hc.files[top.histFileIdx] // O(1) array access per key

---

3. Eliminated SegReaderWrapper

The II segment files store key→txNum-sequence pairs. To read them, the original code wrapped *seg.Reader in a SegReaderWrapper to
satisfy a stream.KV interface:
advance() → stream.KV interface → SegReaderWrapper → seg.ReaderI interface → *seg.Reader
Two interface dispatch layers per call. SegReaderWrapper.Next() also had a redundant HasNext() guard (advance already checked it). Now
ReconItem.g holds *seg.Reader directly:
advance() → *seg.Reader (direct call)
This is why allocations dropped 19% — the SegReaderWrapper heap allocations are gone.

(-5%)

erigon seg check-commitment-hist-at-blk-range got faster: 4m58.038s -> 4m5.294s