feat(planner): price reuse of compatible window pane producers

[zzylol]

Why

Compatible temporal readouts can share a pane producer, but the backend needs a Planner-owned comparison of shared and independent producer costs.

What

Add pane_sharing::select_shared_panes, which selects beneficial groups from executor-certified compatible pane implementations and supplied horizon costs.

How

Group by the executor's complete compatibility key. Charge the longest-retained producer once and retain every consumer's read cost. Reject invalid or overflowing quotes and retain independent implementations when sharing has no benefit.

Before this PR

The optimizer has no costed physical-pane reuse decision for readouts with different lookback windows.

After this PR

A backend can bind sum_over_time(a[1m]) / sum_over_time(a[10m]) to one compatible pane producer while preserving both readout windows. Different sources or incompatible state contracts remain separate.

Evidence

The unit fixture prices independent producers at 12 + 17 = 29; shared production costs 15 + 2 + 2 = 19 and retains the 600-second lookback. A distinct source is excluded.

Screenshots and performance measurements: not applicable; these are modeled costs, not measured speedups.

Verification

• cargo +1.98.0 test -p asap-aware-mapping --lib pane_sharing --offline: 2 passed.
• Tests cover compatibility grouping, longest retention, retained read charges, invalid/overflowing quotes, and no-benefit cases.
• Backend integration and actual readout tests are in fix(planning): unify costed window selection and compatible sum pane sharing ASAPQuery-backend#699, stacked on feat(planning): price the derived window layouts so they can be ranked ASAPQuery-backend#697.
• Formatting and git diff --check pass.

Architectural decisions

Planner owns the cost comparison; the executor supplies physical compatibility and cost evidence. Logical readout windows are not rewritten. This avoids a second optimizer in the backend compiler.

Limitations and follow-up

This reuses compatible selected panes; it does not enumerate all pane widths or jointly reconsider independently selected FullWindow layouts. Equal keys must certify equal unit costs, pane shape and source/state contracts.

Human review — do not complete with an agent

• The MVP boundary is correct.
• New conceptual layers or public interfaces are necessary.
• The before/after description matches the intended product behavior.
• Human reviewer:
• Decision and rationale: