asap-aware-mapping: rebuild applicability.rs findings as a view over the search's candidate-plan space

[zzylol]

Part of #33.

What

Rebuild applicability.rs (PR #247) as a reporting view over the Cascades/Volcano search's resulting candidate-plan space, instead of a parallel system that re-walks the tree itself.

Today, SketchApplicabilityRule/SharedSubexpressionRule each do their own traversal and re-derive a yes/no fact that the search (once the other #33 sub-issues land) will already have computed as part of building the full candidate space. Once that space exists, "which optimizations are applicable" and "which summary is applicable" collapse into the same question: for a given TargetSubDAG, what does its candidate list (across every registered ReplacementStrategy) actually contain?

Design

• ApplicabilityFinding stays the public shape (optimization kind, location, reason) — only its data source changes, from "re-walk the tree" to "read off the search's per-site candidate list."
• OptimizationKind gains one variant per landed ReplacementStrategy (SemanticRewrite, RollupReuse, SharedMultiSubpopulation), same #[non_exhaustive] discipline as today.
• Existing tests in applicability.rs (positive/negative fixtures for sketch and CSE reuse) should keep passing unchanged against the new data source — they're pinning behavior, not implementation.

Depends on

The ReplacementStrategy trait and the search-engine sub-issues landing first; this is the last piece, once there's an actual candidate-plan space to report on instead of re-deriving.

[zzylol]

Related open work

• Follow-up to the candidate-plan search engine in asap-aware-mapping: Cascades/Volcano-style candidate-plan search engine over ReplacementStrategy #252.
• Contributes to the optimizer/applicability umbrella in Add logic to detect which optimizations are applicable to a query workload #33 by making applicability a view of the actual search space instead of a second source of truth.

[zzylol]

Closing as completed after re-reviewing the current implementation.

Evidence:

• 5504a2f implemented the requested PlanSpace-backed view; d459a62 subsequently reframed it as replacement explanation.
• explanation.rs calls search_workload_with and derives findings from PlanSpace/MemoGroup candidate lists through findings_from_plan_space.
• The old ApplicabilityRule, SketchApplicabilityRule, and SharedSubexpressionRule traversal/extension point is deleted; only historical documentation references remain.
• Candidate rationales are reused as the explanation source, so there is no second applicability truth source.
• Deduplication and maximal/shared-site behavior are covered by the explanation tests. cargo test -p asap-aware-mapping explanation --lib passes all 13 tests.

The original design also proposed one reporting enum variant per subsequently landed ReplacementStrategy. Current ExplanationKind intentionally covers sketch approximation and CSE reuse only, while explanation.rs documents semantic rewrites, roll-ups, and other strategies as possible future reporting coverage. That is separable feature coverage, not unfinished PlanSpace migration; it should be tracked independently if dedicated findings are required.