Fix unsound Delta unsigned-widening cast

[joseph-isaacs]

Fixes #8193.

Problem

The Delta cast kernel reused the encoded bases/deltas across primitive-type changes, which is unsound. Delta stores deltas as wrapping_sub at the source bit width and decodes with wrapping_add at the array's bit width, so widening corrupted every value after a delta that wrapped at the source width:

[200u8, 50, 75, 10, 255] -> cast u32 -> [200, 306, 331, 522, 767]   (expected [200, 50, 75, 10, 255])

Two further latent bugs in the same kernel:

• u32 -> f32 passed the guard, entered the reuse path, then errored in Delta::try_new (float dtype) instead of falling back.
• Casting a nullable Delta to non-nullable at the same ptype eagerly rebuilt a non-nullable DeltaArray, silently discarding real nulls.

Fix

Reuse the encoded components only for a nullability change at the same primitive type (adding/keeping nullability, where validity lives in the deltas child). Every ptype change, and dropping nullability, defers to the generic decompress-and-re-encode path, which decodes correctly and validates nulls.

Nothing more is recoverable from metadata: widening would need a wrap/monotonicity flag (value range is insufficient — [0,200] and [200,0] share min/max but only one is safe to widen); narrowing/signedness need the value range, which can't be derived from the components.

Tests

• test_cast_delta_unsigned_widening_wraps — widening corruption regression.
• test_cast_delta_same_width_float_target — u32 -> f32 falls back instead of erroring.
• test_cast_delta_add_nullability — same-ptype add-nullability fast path.
• test_cast_delta_nullability_preserves_nulls — null positions round-trip.
• test_cast_delta_drop_nullability_with_nulls_errors — dropping nullability with real nulls errors.
• existing test_cast_delta_conformance, test_cast_delta_u8_to_u32, test_cast_delta_nullable.

Checks

• cargo test -p vortex-fastlanes — 291 + 3 doctests pass (delta::compute::cast 11/11).
• cargo clippy -p vortex-fastlanes --all-targets --all-features — clean.
• cargo +nightly fmt -p vortex-fastlanes -- --check — clean.

Rebased onto the latest develop and squashed to a single commit.

[codspeed-hq]

Merging this PR will not alter performance

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚡ 1 improved benchmark
❌ 2 regressed benchmarks
✅ 1272 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	cuda/bitpacked_u8/unpack/3bw[100M]	300.2 µs	349.9 µs	-14.19%
❌	Simulation	chunked_varbinview_canonical_into[(100, 100)]	273.1 µs	307.8 µs	-11.28%
⚡	Simulation	chunked_varbinview_opt_canonical_into[(1000, 10)]	225.3 µs	188 µs	+19.87%

Tip

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

---

_{Comparing claude/blissful-ramanujan-7S6HL (cf09b33) with develop (fa2c35a)}