feat: harden WorkTable beta and select index backends

[pathscale]

Summary

This is the consolidated correctness and backend-selection PR for WorkTable 1.0.0-beta.1. It keeps WorkTablesIndex as the default while adding statically selected using backends and hardens the concurrent publication paths exercised by the paper benchmark campaign.

Index backends

• worktables_index remains the default and supports persistence;
• explicit indexset also supports the existing persistence representation;
• explicit congee and arctic require persist: false;
• backend choice is generated statically, without a runtime enum, trait object, or per-operation selection branch; and
• WorkTablesIndex → IndexSet → WorkTablesIndex reload/mutation switching is covered without rebuilding persisted data.

Correctness and stabilization

• exact WorkTablesIndex 0.0.4 and DataBucket 0.5.1 dependency alignment;
• combined backend/versioned-publication regression coverage;
• bounded, feature-gated stable-miss confirmation for B-tree-family providers;
• feature-gated immutable row-version publication and grace-period reclamation;
• reusable persistence-slot coalescing;
• linearized vacuum test oracles;
• same-key upserts hold one generated full-row lock across the existence decision and selected mutation while definitely absent keys retain the optimistic insert path;
• enabled extreme same-key churn regression plus raw insert/delete mutation churn coverage; and
• clean public rustdoc links.

Performance

• successful WorkTablesIndex select delta: +0.461% median across six frozen-binary pairs;
• ordinary update: -1.063% median latency delta;
• absent-key upsert: +0.083% median;
• existing-key upsert: +0.505% median; and
• default latency-sensitive publication behavior remains unchanged unless the stronger modes are explicitly enabled.

Verification

• exact CI command passed locally with the staged release dependencies: cargo test --locked --workspace --all-targets --all-features;
• 129 WorkTable unit tests;
• 334 integration tests passed, 4 explicitly ignored;
• 62 codegen tests;
• benchmark smoke targets passed;
• cargo clippy --locked --workspace --all-targets --all-features -- -D warnings;
• rustdoc passed with warnings denied; and
• the original 240-cell shadow-concurrency campaign passed in full; after the lock-order review rewrite, the stable-confirmation mode passed 120/120 cells while retry-off WorkTablesIndex produced two transient unique-read mismatches in one of 120 cells. Thirty additional high-contention stable-confirmation repetitions passed 30/30. All writer, final-state, primary, stale-hit, torn-row, cardinality, and vacuum counters remained zero, so concurrent-consistency results require the explicit stable-index-read-retry mode.

Dependency order

CI cannot resolve the exact pins until WorkTablesIndex 0.0.4 and then DataBucket 0.5.1 are published. After this PR merges, publish worktable_codegen 1.0.0-beta.1 before worktable 1.0.0-beta.1. Keep this PR in draft until the dependency releases land; the branch itself is ready for review.

The full syntax, persistence contract, backend restrictions, and benchmark plan are in docs/index-backend-dsl-proposal.md.