Faster iterations

[emilk]

Reading a validated column is the hot path: the bounds are known once
(the column length, or a list element's offset range), so re-checking
each element through arrow's value() is wasted work.

Add LogicalType::value_unchecked (safe default = value), overridden
to call arrow's unchecked accessor for the leaf encodings, the trivial
wrappers (Option, newtypes, As, duration, timestamp), and
List/LargeList. Dictionary keeps its value lookup checked, so a
malformed key can't cause UB; Run forwards.

Column (via TypedArray::value_unchecked) and ListValue now read
through it: get/value bounds-check once, and the iterators read
unchecked over their validated range. The iterators also gain
count/last/nth/fold, DoubleEndedIterator, and FusedIterator.

The workspace unsafe_code lint moves from deny to warn, and
quiver_types opts in with a crate-level expect documenting the one
use. Verified with Miri (no UB).

Also move ListValue to its own list_value module.

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

[Copilot]

Pull request overview

This PR optimizes quiver’s validated read hot paths by introducing LogicalType::value_unchecked and routing Column/ListValue iteration through unchecked Arrow accessors after a single upfront bounds check, while also expanding iterator ergonomics and adding a CI Miri soundness job.

Changes:

• Add LogicalType::value_unchecked (defaulting to value) and implement it for key logical types to use Arrow unchecked accessors where sound.
• Update TypedArray, Column iterators, and ListValue to do one bounds check up front and then iterate via unchecked reads; add DoubleEndedIterator/FusedIterator and optimized combinators.
• Add a Miri CI job and adjust workspace unsafe_code linting; move ListValue into a dedicated module and update exports.

Reviewed changes

Copilot reviewed 20 out of 20 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
README.md	Remove “unsafe forbidden” badge to reflect new unsafe usage.
Cargo.toml	Change workspace unsafe_code lint level to warn.
.github/workflows/rust.yml	Add nightly Miri job for soundness validation.
crates/quiver/tests/column.rs	Add tests for new iterator combinators and nested list iteration.
crates/quiver_types/src/datatype.rs	Add LogicalType::value_unchecked and implement it for flat/marker types.
crates/quiver_types/src/typed_array.rs	Route get/value through unchecked access after a single bounds check.
crates/quiver_types/src/column.rs	Optimize iterators to use unchecked reads; add combinators + DoubleEndedIterator/FusedIterator.
crates/quiver_types/src/list.rs	Add list value_unchecked using unchecked offset access; re-export ListValue.
crates/quiver_types/src/list_value.rs	New module for ListValue, updated to iterate via unchecked reads and provide more iterator traits.
crates/quiver_types/src/lib.rs	Add crate-level expect(unsafe_code) and re-export ListValue from new module.
crates/quiver_types/src/option.rs	Implement value_unchecked for Option<L>.
crates/quiver_types/src/newtype.rs	Forward value_unchecked through newtype wrappers.
crates/quiver_types/src/duration.rs	Forward value_unchecked to Arrow unchecked accessor.
crates/quiver_types/src/timestamp.rs	Forward value_unchecked to Arrow unchecked accessor.
crates/quiver_types/src/string.rs	Implement unchecked access for AnyUtf8 encodings.
crates/quiver_types/src/binary.rs	Implement unchecked access for binary encodings and AnyBinary.
crates/quiver_types/src/fixed_size_binary.rs	Implement unchecked access for fixed-size binary.
crates/quiver_types/src/dictionary.rs	Implement unchecked key read but keep checked value lookup to avoid UB on malformed keys.
crates/quiver_types/src/run.rs	Forward unchecked reads through run-end mapping.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.