Fix flaky RSpec: include local_hcb_code.ledger_item in calculated_ledger_item

[Copilot]

Summary of the problem

bad_settled_mapping_spec.rb fails non-deterministically (reproducible with --seed 17199) with:

RuntimeError: CanonicalTransaction 4 has calculated a different ledger item from its local_hcb_code. ( vs. 6)

CanonicalTransaction and CanonicalPendingTransaction each use HCB-000-{id} as their hcb_code when no linked object exists. Because they draw from independent PostgreSQL sequences, they can coincidentally receive the same id. When that happens, the CPT's assign_ledger_item creates a Ledger::Item and attaches it to the shared HcbCode. When the CT is then created, its assign_ledger_item check sees calculated_ledger_item = nil but local_hcb_code.ledger_item = <CPT's item> and raises via Rails.error.unexpected.

Describe your changes

Added local_hcb_code&.ledger_item as a third fallback in CanonicalTransaction#calculated_ledger_item:

# before
@calculated_ledger_item ||= Ledger::Item.find_by(short_code:) || linked_object_v2&.ledger_item

# after
@calculated_ledger_item ||= Ledger::Item.find_by(short_code:) || linked_object_v2&.ledger_item || local_hcb_code&.ledger_item

This makes CT consistent with CanonicalPendingTransaction#assign_ledger_item, which already uses local_hcb_code.ledger_item as its primary source. The Rails.error.unexpected check is preserved and still fires for genuine mismatches—when short_code or linked_object_v2 resolves to a specific ledger item that disagrees with what the HcbCode stores. The false positive (nil vs. an already-assigned item from a co-grouped record) is eliminated.

[garyhtou]

@copilot My intuition tells me that this is not the right fix because we do not want CanonicalTransaction to depend on HCBCode. We plan to deprecate HCBCode since Ledger::Item is replacing it.

[Copilot]

@copilot My intuition tells me that this is not the right fix because we do not want CanonicalTransaction to depend on HCBCode. We plan to deprecate HCBCode since Ledger::Item is replacing it.

Good point. I've reverted the calculated_ledger_item change so it no longer touches local_hcb_code. Instead, the fix is in the assign_ledger_item check: the mismatch error now only fires when calculated_ledger_item is present but disagrees with local_hcb_code.ledger_item. The false positive (nil vs. an already-assigned item) is eliminated without adding any new dependency on HcbCode to calculated_ledger_item.