test(early-cashout): add dust stake edge-case tests for #407

[diahtech1]

Summary

Adds a comprehensive test suite for adversarial dust stake interactions with early cash-out and fee paths. Pins down the exact conservation invariant for every combination of dust-vs-non-dust stakes and fee-on / fee-off settlement.

Closes #407

---

Motivation

Adversarial dust stakes (1-stroop positions) and their interaction with the fee-on / fee-off paths can break token conservation. When stake * penalty_bps / BPS_DENOMINATOR floors to zero, the user receives a full refund with no forfeit — the pool shrinks by the full stake while zero goes to treasury. Until now, there were no tests verifying that this path maintains the conservation identity:

sum(cashouts) + sum(settlement_payouts) + protocol_fee_delta == original_total_pot

---

What's included

13 scenario tests + 1 property-based test

#	Test	Scenario
1	dust_cashout_fee_off_full_refund	Single 1-stroop dust stake cash-out, fee off → full refund
2	dust_cashout_fee_on_full_refund	Single 1-stroop dust stake cash-out, fee on → full refund
3	dust_cashout_fee_on_forfeit_rounds_to_zero	Stake=9999, penalty=1 bps → forfeit floors to 0
4	dust_cashout_settle_remaining_fee_off	Dust cash-out + remaining pool settles, fee off
5	dust_cashout_settle_remaining_fee_on	Dust cash-out + remaining pool settles, fee on
6	multiple_dust_cashouts_fee_off	Two 1-stroop participants both cash out
7	multiple_dust_cashouts_settle_fee_on	Dust cash-out + settle with 5% fee
8	dust_cashout_pool_totals_consistency	Pool sums decrease by exactly the dust stake
9	all_dust_cashouts_fee_on_full_round	Every participant is dust, all cash out → pool drains to zero
10	dust_on_loser_side_cashout_fee_on	Dust on eventual losing side cashes out
11	dust_forfeit_boundary_transition	Smallest stake where forfeit becomes non-zero (stake=10, penalty=10%)
12	dust_cashout_fee_on_winnings_model	Dust with FeeOnWinnings model
13	dust_cashout_max_penalty_still_full_refund	Max penalty (10%) × min stake (1 stroop) → still full refund

Property-based test (proptest, 25 cases)

Verifies conservation for random dust_stake in [1..99], penalty_bps in [1000..10000], and arbitrary opposing stake. Asserts:

• cashout + forfeit == stake
• pool_up == 0 after cash-out
• Full round conservation: cashout + settlement_payout + treasury_delta == original_pot

---

Key edge cases covered

• Forfeit rounding: stake * penalty_bps / 10_000 floors to zero for any stake < 10_000 / penalty_bps. This means 1-stroop dust always gets a full refund regardless of penalty rate.
• Boundary transition: The exact stake where forfeit transitions from 0 to 1 (stake=10 with 10% penalty → forfeit=1, cashout=9).
• One-sided settlement after dust: When all Up bettors cash out, the remaining Down pool is one-sided → refund with no fee.
• Fee model interaction: Both FeeOnPot and FeeOnWinnings behave correctly when the dust cash-out precedes settlement.

---

Files changed

• contracts/src/tests/early_cashout_dust.rs — new (854 lines)
• contracts/src/tests/mod.rs — added mod early_cashout_dust;

---

Testing

All tests exercise the contract through the public Soroban client interface (no internal state manipulation beyond the setup helpers for fee/penalty configuration). Each test verifies conservation with explicit assert_eq! on the full accounting identity.

Note: The Rust toolchain was unavailable in the CI environment at time of writing. The tests are designed to be run with cargo test -p contracts early_cashout_dust and should be validated in CI.

---

Checklist

• Tests cover dust (1 stroop) with fee on and off
• Tests cover the forfeit rounding-to-zero path
• Tests cover the forfeit boundary (0 → 1 transition)
• Tests verify pool totals decrease by exactly the stake
• Tests verify conservation identity: cashouts + payouts + treasury == original pot
• Property-based test covers the full dust regime
• Both FeeOnPot and FeeOnWinnings models tested
• CI passes (awaiting toolchain availability)