feat(post-asap): define summary maintenance modes - #300

Current-main integration review: this PR is not ready to merge as-is after #295 and #299. The conflict is architectural rather than mechanical. #295 now owns the canonical recurrence vocabulary (recurrence::{CostRate, EvaluationRate, UpdateRate, Horizon}) and recurrence-aware selection APIs, while this branch independently introduces duplicate CostRate/EvaluationRate types and a second recurring-cost path in cost_model.rs. #299 now also owns nested-summary legality through ResultGuarantee/AccuracyModel; phase validation must compose with that model rather than independently rejecting or admitting the same nested candidates. Please rebase on current main, reuse the recurrence types and profiles instead of defining parallel units, and make exact composition preserve/propagate the guarantees introduced by #299. The merged implementation should retain the conservative missing-statistics fallback and phase tests from this PR.

Implemented the workload model and summary lifecycle maintenance from #310 on top of this PR's phase-aware exact composition.

Highlights:

• normalized batch + repeated query workloads with recurrence, predictability, independent time selection, accuracy/latency requirements, and uncertain demand
• separate data workload with arrival, fresh evidence, cardinality, and DataDistribution
• workload-derived recurrence/update profiles and accuracy evidence
• explicit ephemeral/prepared/shared/continuously-maintained lifecycle alternatives
• horizon-normalized lifecycle costs, capability/phase validation, assumptions, and structured rejection reasons
• materialize_with_lifecycles bridge from this PR's GlobalSelection to plan + state deployments

Validation: workspace check, workload/lifecycle/accuracy tests, all PromQL and SQL frontend tests, and all 11 exact_composition integration tests pass.

Addressed the four follow-up review findings in e117dcb:

• lifecycle planning now takes an explicit WorkloadDemand binding, so unrelated workload entries cannot create false reuse for a target
• recurrence profiles require an explicit root-to-workload-entry index mapping instead of assuming batch/repeating container order
• scheduled evaluation rates count only executions inside the optimization horizon
• future-dated data evidence and demand estimates are not considered fresh

Added regression coverage for all four cases. Local validation passes: 5 workload tests, all 237 asap-aware-mapping tests, all 11 exact-composition integration tests, workspace strict Clippy, rustfmt, and diff checks.

Addressed the second code-and-doc review in ec714b6:

• lifecycle-aware materialization now compares aggregate summary deployment cost with raw recomputation and can fall back to KeepPreAsap
• per-summary incremental-update/delete capabilities gate maintained lifecycles; moving real-time windows require deletion support
• prepared state is legal only when every bound consumer is predictable and covered by the declared window
• empty and duplicate WorkloadDemand bindings are rejected
• docs now distinguish implemented latency/provenance behavior from future integrations, update the stale glossary-gap table, and state the remaining broader sibling-candidate search boundary

Regression coverage was added for raw fallback, unsupported deletion, mixed predictable/ad-hoc consumers, and invalid bindings. Local strict Clippy, rustfmt, 241 mapping tests, 5 workload tests, and 11 exact-composition integration tests pass.

@milindsrivastava1997 Splitting the PRs, FYI.