fix: coredump when sequence field is part of primary key

[lxy-9602]

Purpose

Problem Description

In the MergeFileSplitRead logic for PK MOR tables:

1. When the sequence.field is included in the primary key fields, the current implementation first splits the primary key (PK) and non-PK fields based on user read schema.
2. After splitting, the sequence.field is added to the non-PK fields. However, this results in an incorrect arrow::Schema that contains duplicate fields (e.g., sequence.field appears in both PK and non-PK fields).
3. The arrow::Schema::GetFieldIndex() function does not handle duplicate fields gracefully and returns -1 when duplicates are detected. This leads to a coredump during execution.

---

Solution

To resolve this issue:

• Modified the order of operations in MergeFileSplitRead:
  • First, add the sequence.field to the appropriate field set.
  • Then, split the PK and non-PK fields.

This ensure no duplicate fields are introduced in the arrow::Schema.

Tests

WriteAndReadInteTest.TestPKWithSequenceFieldInPKField
WriteAndReadInteTest.TestPKWithSequenceFieldPartialInPKField