Relocate Delta's selection policy into cost-model priors; plumb the model through the builders

[connortsui20]

Tracking Issue: #8701

Rationale for this change

Delta's selection policy — the 0.95 "delta tax" and the 1.25 minimum-win floor — is an execution-speed judgment expressed as ratio fudge factors inside the scheme. This PR moves that policy onto the cost model as data (a per-scheme prior), and makes the model configurable end to end: through BtrBlocksCompressorBuilder and, via the file writer's strategy, into both the data and stats compressors (#7697).

Stacked on #8699. Behavior-preserving by construction: the model applies the identical arithmetic in the identical order, so scores are bit-identical — zero golden snapshot churn, including the variant where Delta wins corpus entries.

What changes are included in this PR?

• SizeCost gains per-scheme priors (SchemePrior { multiplier, min_ratio }); DeltaScheme reports raw ratios, and its prior is registered on the btrblocks default cost model, where the policy is documented as policy, not measurement.
• BtrBlocksCompressorBuilder::with_cost_model; a configured model rides through WriteStrategyBuilder into both writer compressors. Liveness tests prove the plumbing drives selection (an inverted model flips the winner, and writes a strictly larger file).

What APIs are changed? Are there any user-facing changes?

Additive: SchemePrior, SizeCost::with_scheme_prior, BtrBlocksCompressorBuilder::with_cost_model, and a cost re-export from vortex-btrblocks. Deprecated: DeltaScheme::new(min_ratio) — the floor belongs on the model; the constructor keeps working and can only tighten policy. Embedders assembling CascadingCompressor directly with DeltaScheme should attach a prior-bearing model (as the builder does) to retain the handicap. Default compressor output is byte-identical.

[codspeed-hq]

Merging this PR will not alter performance

⚡ 2 improved benchmarks
❌ 3 regressed benchmarks
✅ 1616 untouched benchmarks
🆕 39 new benchmarks
⏩ 47 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	chunked_varbinview_into_canonical[(100, 100)]	271.6 µs	327.8 µs	-17.13%
❌	Simulation	chunked_varbinview_opt_canonical_into[(100, 100)]	305.3 µs	367.6 µs	-16.96%
❌	Simulation	encode_primitives[i64, (1000, 512)]	117.5 µs	137.6 µs	-14.6%
⚡	Simulation	chunked_bool_canonical_into[(1000, 10)]	16.4 µs	14.6 µs	+12.36%
⚡	Simulation	true_count_vortex_buffer[128]	638.3 ns	580 ns	+10.06%
🆕	Simulation	ilike_contains	N/A	782.5 µs	N/A
🆕	Simulation	like_contains	N/A	154.5 µs	N/A
🆕	Simulation	like_exact	N/A	43.9 µs	N/A
🆕	Simulation	like_per_row_patterns	N/A	168.7 µs	N/A
🆕	Simulation	like_prefix	N/A	50.9 µs	N/A
🆕	Simulation	like_regex	N/A	623 µs	N/A
🆕	Simulation	like_suffix	N/A	64.9 µs	N/A
🆕	Simulation	cached_indices_i128[0.01]	N/A	29 µs	N/A
🆕	Simulation	cached_indices_i128[0.1]	N/A	82.1 µs	N/A
🆕	Simulation	cached_indices_i32[0.01]	N/A	22.5 µs	N/A
🆕	Simulation	cached_indices_i32[0.1]	N/A	55.9 µs	N/A
🆕	Simulation	patterns_i128[Contiguous]	N/A	234.8 µs	N/A
🆕	Simulation	patterns_i128[Random]	N/A	290.4 µs	N/A
🆕	Simulation	patterns_i128[Runs]	N/A	244.8 µs	N/A
🆕	Simulation	random_i128[0.01]	N/A	29.2 µs	N/A
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Tip

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

---

_{Comparing cost-model-r5-builders (a89ad4c) with cost-model-r3-lower-bound (d7a3b3a)²}

Footnotes

1. 47 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 cost-model-r3-lower-bound (02309ed) during the generation of this report, so afb6425 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩