fix: execution fee refund, limit order boundary tests, weighted avg entry price, write-ordering docs

[kulayddon]

Summary

• exchange_router: cancel_order does not refund execution fee to user #258 exchange_router / order_handler: cancel_order now refunds the full execution_fee to the user in addition to collateral. Previously, the fee was silently retained in the order vault even though no keeper performed any work. A matching test (cancel_order_refunds_execution_fee_to_user) verifies both collateral and fee are returned and the vault balance reaches zero.

• order_handler: limit order incorrectly triggers when price equals trigger_price exactly #259 order_handler: all limit/stop trigger conditions already use inclusive bounds (<= / >=), so LimitIncrease with oracle_price == trigger_price always executed. Added boundary tests for all four relevant order-type/direction combinations (limit_increase_at_trigger_price_executes, limit_decrease_long_at_trigger_price_executes, limit_decrease_short_at_trigger_price_executes, stop_loss_long_at_trigger_price_executes) to lock in this behaviour and prevent regressions.

• reader: get_position_info returns wrong entry_price after multiple position increases #260 reader / libs/types: added avg_entry_price: i128 to PositionInfo. Both get_position_info and get_position_info_by_key now compute the weighted average entry price as size_in_usd / size_in_tokens × TOKEN_PRECISION. Also fixed a pre-existing compile error: IncreasePositionParams was missing the for_positive_impact field added in issue feat: maker/taker fee tiers — lower fees for limit orders, higher for market orders #284. A test (position_avg_entry_price_is_weighted_average_after_two_increases) opens a position twice at different prices and asserts the implied average lies between the two prices.

• data_store: no write guard — two handlers writing same key in one tx produce inconsistent state #261 data_store: added a doc comment on apply_delta_to_u128 explaining that Soroban's sequential execution model within a transaction makes a separate version/nonce guard unnecessary. True concurrent writes across transactions are prevented by Soroban's footprint model. A test (sequential_pool_amount_writes_both_accumulate) demonstrates that two consecutive apply_delta_to_u128 calls accumulate correctly.

Test plan

• cancel_order_refunds_execution_fee_to_user — collateral + fee returned, vault empty
• limit_increase_at_trigger_price_executes — boundary execution for LimitIncrease
• limit_decrease_long_at_trigger_price_executes — boundary execution for long take-profit
• limit_decrease_short_at_trigger_price_executes — boundary execution for short take-profit
• stop_loss_long_at_trigger_price_executes — boundary execution for long stop-loss
• position_avg_entry_price_is_weighted_average_after_two_increases — weighted avg correctness
• sequential_pool_amount_writes_both_accumulate — sequential write ordering

Closes #258
Closes #259
Closes #260
Closes #261

[drips-wave]

@kulayddon Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits