perf(reader): Implement AsyncFileReader get_byte_ranges and coalesce close ranges

[mbutrovich]

Which issue does this PR close?

• Partially address Epic: ArrowReader performance improvements for split FileScanTasks #2172.

What changes are included in this PR?

• Adapt range coalescing from object_store.

Are these changes tested?

Existing tests, some new ones just to sanity check merge_ranges. Also ran full Iceberg Java suite via Comet. Benchmarks below.

[mbutrovich]

I ran SELECT SUM(l_extendedprice * l_discount) on a SF100 lineitem TPC-H table accessed via S3 (essentially just the scan from query 6 in TPC-H) with Comet. Here's the breakdown:

Metric	main (uses get_bytes)	PR #2181 (uses get_byte_ranges)	Change
Total GETs	773	500	35% fewer
Total bytes	3,535 MB	3,535 MB	same

Percentile	main (uses get_bytes)	PR #2181 (uses get_byte_ranges)	Change
P50 (median)	1.5 MB	3.6 MB	2.4x larger
P75	3.3 MB	17.0 MB	5.2x larger