feat(fusio-manifest): optimize scan range path and split benches

[ethe]

Intent

This PR reduces scan_range read amplification in fusio-manifest by pruning non-overlapping data earlier (segment/run/block), and separates benchmark
responsibilities so point-get and range-scan performance can be measured independently and consistently.

Benefits

1. Lower read amplification for range queries, especially when query windows are narrow.
2. Fewer and smaller checkpoint reads (range reads instead of full payload reads on scan path).
3. Clearer benchmark signal by separating point-get and scan-range workloads.

Measured impact (local disk benchmark)

• L1-ScanRangeHit: -88.91% latency (~9.01x faster)
• L1-ScanRangeInRangeMiss: -89.13% latency (~9.20x faster)
• L1-ScanRangeMiss: -89.51% latency (~9.53x faster)
• MultiLevel-ScanRangeHit: -59.62% latency (~2.48x faster)
• MultiLevel-ScanRangeInRangeMiss: -59.35% latency (~2.46x faster)
• MultiLevel-ScanRangeMiss: -58.97% latency (~2.44x faster)