Phase 5: bounded LongBridge offline research conversion

[Pigbibi]

Scope

Adds an isolated, pure LongBridge offline research conversion for the frozen QQQM/TQQQ/BOXX Phase 4 v3 candidate. The input is the original frozen plan plus an already funded whole-share research proposal. The adapter independently verifies candidate/time identity, target shares, funding order, actual fill quantities, costs, cash, member reserve and account identity before producing LongBridge-shaped offline rows.

Only three new research/test/documentation files are changed. No production execution, account, market-data, release, risk, workflow or broker-submission path is modified.

Evidence

• 16 synthetic and negative direct tests pass. They cover conversion fidelity, whole-share output, zero-share omission, deterministic repeat evaluation, and rejection of tampered funded inputs.
• Approved local restricted-input check: eight predeclared first-open Phase 5 paths at 10 bps match the separately saved frozen research ledger in shares, settled cash, executed cost and TQQQ member reserve. One funding shortfall remains explicit. No raw private data is committed or uploaded.
• Ruff and staged whitespace checks pass.
• Independent Astra incremental review: GO for this bounded conversion and first-open accounting checks. Astra did not access private raw data.

Boundaries

This is conversion fidelity for an already computed funded research proposal, not a second strategy funding engine or a native LongBridge execution test. The current native paper value-target mapper produced fractional quantities and did not match any of the eight whole-share sets; this PR does not change that mapper. Existing holdings, pending sales, receivables, collateral/options, full-window settlement and real broker behavior remain outside the supported slice. It grants no paper, shadow or live authority, and requests no merge or deployment.