Refactor read driver as an explicit stream

[joseph-isaacs]

Summary

Stacked on #9384.

• replace the nested async read-driver loop with an explicit ReadDriver<R>: Stream state machine
• replace the nested async_stream result adapter with an explicit ReadRangeResults state machine
• drain every immediately-ready completion and refill freed read slots from IoRequestStream before yielding
• preserve the existing physical concurrency cap, validation, unknown/missing-result handling, metrics, tracing, and panic propagation

ReadRangeResults has explicit Reading and Missing states: it matches out-of-order range results while reading, then reports every unresolved request if the underlying stream ends early.

When several results become ready together, the driver now resolves all of them and submits one refill group for the freed slots in the same poll. It does not add another FileReadAt or ObjectStore queue.

Validation

• cargo +nightly fmt --all -- --check
• cargo test -p vortex-file (142 passed; 1 doctest ignored)
• cargo clippy -p vortex-file --all-targets --all-features -- -D warnings
• RUSTDOCFLAGS='-D warnings --cfg docsrs' cargo doc --profile ci --no-deps --document-private-items -p vortex-file
• deterministic refill test verifies three simultaneous completions produce one three-range refill while retaining the configured active-read cap

[codspeed-hq]

Merging this PR will regress 1 benchmark

⚡ 4 improved benchmarks
❌ 1 regressed benchmark
✅ 1957 untouched benchmarks
⏩ 121 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	optimize_or_chain[200]	546.8 µs	1,177.4 µs	-53.56%
⚡	Simulation	compact_sliced[(16384, 10)]	2 ms	1.5 ms	+37.48%
⚡	Simulation	compact_sliced[(4096, 10)]	528.6 µs	391.6 µs	+34.99%
⚡	Simulation	take[small_m/shuffled/primitive/nonnull/chunks=16384/indices=16]	1.2 ms	1 ms	+13.21%
⚡	Simulation	eq_pushdown_low_match	1.6 ms	1.4 ms	+11.06%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing ji/read-ranges-state-machine (c3f5d45) with ji/read-ranges (0a48a0c)}

Footnotes

1. 121 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩