(OPT) Reuse certified first-head PLAN slices instead of rebuilding the right cursor view #902

Follow-up to #900 and #899: avoid rebuilding the right first-head segment view when the nested left first-head continuation already holds the same immutable segment. This targets singleton segments too. Discovery is source-only at local candidate d458d05 (production unchanged from 84432); no implementation, application admission or net gain is established.

In comparison_owner/paired_value.rs, preserve both preparations and every original left/right work event. In the fresh nonempty left-head continuation, pop the right link at its original boundary and compare the two lexical indices before indexing the right arena or constructing its slice/head/rest views. Equal indices in the same prepared immutable arena prove a nonempty identical segment: assign right_items from the already retained left post-head slice, advance the right item and enter #900's certified-head continuation. Foreign indices, loaded segments, exhaustion and tails retain complete ordinary behavior. No persistent certificate, pointer-content equivalence, preliminary walk or new authority.

Derive the final invoice from actual authored typed roles. Discovery layout: original right STEP+EMPTY=108 and LOAD=146; proposed BASE=36, POP=42, certified RHS=48 (including the retained 24-byte conservative as_ref invoice), foreign VIEW_HEAD=176 and loaded/exhausted HEAD=72. Thus ordinary branches remain fully priced, while a certificate removes an additional right arena/slice/head/rest packet beyond #900. A straightforward implementation adds four distinct primitive payment sites, acquired conservatively even when cold. Conditional complete-run equation on that layout: old900 minus successor = 128J - 4PAYMENT_HEADERM - 17(Qnew-Qold), before any further authored roles. Never add #900's separate 81*J or credit its as_ref roles twice. Foreign empty right segments can increase Q; refused prefixes require actual acquired-packet counts. These are a proposed component equation, not a projected application saving.

Splitting STEP changes real funding order: at the local boundary, 107 physical bytes and zero work may produce old physical refusal versus new work refusal. Specify that boundary honestly; preserve sticky failure and all six diagnostic fields at actual new sites. Do not pay a phantom old bank or assert unchanged mixed-budget priority. With sufficient physical funding, retain every original work cut, semantic result, ordering and heapsort swap.

Acceptance:

• Freeze and independently compile the complete immediate predecessor, all macros and invoice; retain every older source/oracle/control. Couple new children to actual source unions using original classification tools.
• Prove same-index first-head provenance, preparation/malformed-link/depth behavior and immutable same-invocation borrowing; preserve full independent final PLAN/HIR verification and both representations.
• Derive all BASE/POP/HEAD/RHS, receiver/control/assignment/drop and cold-site invoices from final code. Census acquired versus completed certificates, Q events, fallback/empty cases and work; preserve surviving conservative fees, caps and hostile controls.
• Exercise same/different IDs, equal contents in foreign IDs, empty/singleton/long segments, asymmetric entry, invalid histories, every work cut, exact/short/zero/floor/sticky/foreign/restored budgets, mixed byte/work cuts, post-swap behavior and complete allocation/reallocation/Drop.
• Run the unchanged whole32 and real application owners at 67,108,864 bytes plus relevant focused checks and scoped production Clippy in a consolidated batch. Never full gates. Keep parents open where acceptance remains missing and distinguish shipping, synthetic and refused-prefix evidence.

Measurement attribution: the old352 trace with 62,710 rounds and 8,493 identity hits is rows.main in the whole32 fixture, not the normalizer application and not current J/G/Q. The separate old normalizer's committed route prefixes are not removable allocations. Neither proves current application ranking, token efficiency or billed-cost advantage.