fix(cancel_refund): zero entry.net and decrement market totals on refund

[esthereze]

Summary

Fixes the inconsistency reported in #58 where cancel_refund zeroed entry.gross but left entry.net intact, causing get_bet to still report a non-zero staked amount after a refund, and market total_yes/total_no to remain inflated.

Root cause

cancel_refund only did:

let gross = entry.gross;
entry.gross = 0; // idempotency guard

This left entry.net untouched, so:

• get_bet returned amount: e.net (non-zero) even after the bet was refunded
• market.total_yes/total_no were never decremented, so market totals were inconsistent with the refunded bet state

Fix

BetEntry already carries both is_yes and net, so no storage-model change is required. The fix is entirely localised to cancel_refund:

let gross = entry.gross;
let net = entry.net;
// Issue #58: zero both gross (idempotency guard) and net
entry.gross = 0;
entry.net = 0;
env.storage().persistent().set(&bet_key, &entry);

// Issue #58: decrement market totals
if entry.is_yes {
    market.total_yes = market.total_yes.saturating_sub(net);
} else {
    market.total_no = market.total_no.saturating_sub(net);
}
env.storage().persistent().set(&mkt_key, &market);

The issue description claimed this was unsolvable without a storage-model change, but that is not the case — per-bet side (is_yes) and net amount are already stored on BetEntry, making the fix purely local. saturating_sub is used defensively to prevent underflow.

Changes

• prediction_market/src/lib.rs: cancel_refund now zeroes entry.net and decrements the appropriate market total
• prediction_market/src/tests.rs: New test test_cancel_refund_clears_net_and_market_totals verifies both invariants

Testing

New test test_cancel_refund_clears_net_and_market_totals covers:

1. get_bet(amount) returns 0 after cancel_refund (was non-zero before fix)
2. market.total_yes and market.total_no are both 0 after all bettors have been refunded (were inflated before fix)
3. get_bet_gross still returns 0 (existing idempotency guard unchanged)

Closes #58