fix: price delta's bases and tail padding in its compression estimate

[rapour]

Summary

Fixes: #9601

Changes

DeltaScheme estimated full_width / delta_bits * DELTA_PENALTY, the cost of the
residual payload alone. Delta also stores a bases child and pads its deltas child up
to whole 1,024-value chunks, and neither term was in the estimate, so the scheme can
win a column and then lose it on real bytes.

Both are cheap to price exactly:

• FastLanes keeps one base per lane per chunk, and a chunk of W-bit values has
  1024/W lanes, so each chunk's bases occupy 1024 bits whatever W is: one bit per
  stored element.

• The deltas child is len.next_multiple_of(1024) long, so a column that overshoots a
  chunk boundary pays for residuals it stores no values for.

  len * full_width / (padded_len * (delta_bits + 1)) * DELTA_PENALTY
  

The same helper now supplies the best_so_far early-exit bound since Delta always pays for the bases.

API Changes

N/A

[codspeed-hq]

Merging this PR will regress 2 benchmarks

⚠️ 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.

⚡ 3 improved benchmarks
❌ 2 regressed benchmarks
✅ 2092 untouched benchmarks
⏩ 106 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	cold_misaligned[(16, 64)]	344.6 µs	390.2 µs	-11.67%
❌	Simulation	compress_fsst[(500, 64, 8)]	480.4 µs	538 µs	-10.72%
⚡	WallTime	arrow_checked_add_u32_neon[16384]	20.4 µs	13.5 µs	+50.77%
⚡	WallTime	arrow_checked_add_u32_avx512[16384]	21.4 µs	17.6 µs	+21.29%
⚡	WallTime	arrow_checked_add_u32_avx2[16384]	21.3 µs	17.7 µs	+20.38%

Tip

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

---

_{Comparing rapour:fix/delta-estimate (31a5290) with develop (17bd9e2)}

Footnotes

1. 106 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. ↩