fix(layout): don't panic collecting an empty stream

[miniex]

Summary

CollectStrategy collects its whole input into a single chunk for a child that requires exactly one chunk. When the input stream is empty, e.g. writing a zero-row table with a nullable struct column whose validity substream is empty, no chunk supplied a sequence id and it panicked with must have visited at least one chunk. CollectStrategy now yields nothing on empty input, and FlatLayoutStrategy returns an empty ChunkedLayout instead of requiring a single chunk, so no segment is written for an empty array.

The issue mentions the fuzzer's assume() guard could be dropped once this is fixed. I left it in place here: reading a nullable struct nested in a struct back is a separate bug (#8348), so the round-trip only works once both are fixed.

Closes: #8347

Testing

Added a regression test in vortex-file/tests/test_write_table.rs that writes a zero-row nullable struct column; it panicked before this change and now writes and reads back as zero rows. The issue's Python repro (vx.io.write of an empty struct column) no longer panics. cargo nextest run -p vortex-layout -p vortex-file passes (177 tests, including the segment-ordering tests). fmt --all + clippy --all-targets --all-features clean.

---

I'm Korean, so sorry if any wording reads a little awkward.

[codspeed-hq]

Merging this PR will degrade performance by 25.57%

⚡ 3 improved benchmarks
❌ 13 regressed benchmarks
✅ 1493 untouched benchmarks
⏩ 83 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	slice_empty_vortex	368.3 ns	2,628.6 ns	-85.99%
❌	Simulation	slice_vortex_buffer[1024]	871.4 ns	1,335 ns	-34.73%
❌	Simulation	slice_vortex_buffer[16384]	871.4 ns	1,335 ns	-34.73%
❌	Simulation	slice_vortex_buffer[2048]	871.4 ns	1,335 ns	-34.73%
❌	Simulation	slice_vortex_buffer[128]	871.4 ns	1,335 ns	-34.73%
❌	Simulation	slice_vortex_buffer[65536]	871.4 ns	1,335 ns	-34.73%
❌	Simulation	chunked_varbinview_canonical_into[(1000, 10)]	162.2 µs	198 µs	-18.08%
❌	Simulation	chunked_varbinview_into_canonical[(1000, 10)]	177.7 µs	213.5 µs	-16.78%
❌	Simulation	search_index_below_min_chunked	1.3 ms	1.5 ms	-13.61%
❌	Simulation	search_index_mixed_out_of_range_chunked	1.3 ms	1.5 ms	-13.31%
❌	Simulation	count_i32_clustered_nulls	47 µs	53.8 µs	-12.68%
❌	Simulation	search_index_full_range_random_chunked	1.4 ms	1.6 ms	-12.06%
❌	Simulation	chunked_varbinview_canonical_into[(100, 100)]	273.2 µs	307.9 µs	-11.27%
⚡	Simulation	chunked_varbinview_opt_into_canonical[(1000, 10)]	229.1 µs	192.7 µs	+18.86%
⚡	Simulation	bitwise_not_vortex_buffer_mut[128]	215.3 ns	186.1 ns	+15.67%
⚡	Simulation	bitwise_not_vortex_buffer_mut[1024]	275.6 ns	246.4 ns	+11.84%

Tip

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

---

_{Comparing miniex:fix/empty-struct-write-panic (174774e) with develop (85aad72)²}

Footnotes

1. 83 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. ↩

2. No successful run was found on develop (0ed06b3) during the generation of this report, so 85aad72 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[onursatici]

Thanks for the contribution!
This seems correct, but I think we should fix this on a lower level.
We generally shouldn't use eof if we do not intend the chunk to go towards the end of file, so although it makes it possible to write an empty chunk, I think the right behaviour of collect strategy is to not yield anything on empty input.
If you just make that change the empty stream would propagate to the children and the flat layout would then fail. I think in the flat layout if there is no chunk we should return an empty ChunkedLayout which allows for no segments to exist at all, unlike flat layout which enforces one segment. If we do this we can save from writing a segment just to store an empty array

So the shape would be:

• CollectStrategy: empty input yields no chunks
• FlatLayoutStrategy: empty input returns a zero-row, zero-child ChunkedLayout
• non-empty flat input keeps the existing exactly-one-chunk behavior

[miniex]

Thanks for the steer - done.

CollectStrategy now yields nothing on empty input, and FlatLayoutStrategy returns an empty ChunkedLayout instead of requiring a single chunk, so no segment is written for an empty array. Dropped the eof minting as you suggested. Non-empty flat input keeps the exactly-one-chunk path.

Verified the empty file still opens and scans back to zero rows; full vortex-layout + vortex-file suites green.