PrepaymentABS: SMM can exceed 1 before exhaustion, over-reporting prepayment vs outstanding balance #358

Context — the PrepaymentABS feature from PR #322 (#323). The crash and negative-SMM failure modes at/after the implied balance-exhaustion month are fixed in src/Asset.hs (smm now returns 0 when 1 - r*(m-1) <= 0).

Residual issue — before exhaustion, SMM(m) = r / (1 - r*(m-1)) can exceed 1:

• r = 6%, m = 17 → 1.5 (150% of the outstanding balance in one month)
• r = 2%, m = 50 → 1.0
• In general, the month just before exhaustion (m = floor(1/r) + 1) already yields SMM >= 1; any projection that runs to ~1/r months (e.g. a 30-year monthly schedule with r >= ~0.28%/month) enters this region.

Downstream impact — the rate is applied without clamping the amount:

• src/AssetClass/Mortgage.hs:223 — newPrepay = mulBR (max 0 (begBal - newDefault)) ppyRate
• same pattern at src/AssetClass/Mortgage.hs:53, :274; src/AssetClass/Loan.hs:51; src/AssetClass/Installment.hs:55

The balance is clamped (intBal = max 0 (begBal - newDefault - newPrepay)), but the recorded newPrepay row can exceed begBal - newDefault — i.e. reported prepayments exceed the outstanding balance, row identities (begin balance vs outflows) stop tying, and pool aggregations over-report prepayments.

Suggested fixes (decision needed)

1. (recommended) Clamp the amount at the application sites: newPrepay = min (begBal - newDefault) (mulBR (max 0 (begBal - newDefault)) ppyRate) — preserves the modeled rate, keeps row identities consistent, and applies uniformly to every prepayment source.
2. Clamp the rate to min 1 inside smm — simpler, but changes the ABS model semantics (the final months effectively prepay 100% of the remaining balance).
3. Reject rate/horizon combinations that cross exhaustion with a descriptive Left — consistent with the other buildPrepayRates validations, but reduces usability.

Acceptance criteria

• With PrepaymentABS 6%, month age 17: newPrepay <= begBal - newDefault, and the row satisfies begBal = endBal + principal + prepayment + default.
• A unit test pinning the chosen semantics (Mortgage/Loan/Installment projection).

References — PR #322, review finding H1; issue #323 (feature); the exhaustion fix in src/Asset.hs.