Complete SExI SQL assessment with pytest coverage

[bogdanalexb]

Summary

• Rebuild the EMPLOYEE, EXPENSE, SUPPLIER, and INVOICE tables from the supplied source files with exact Trino types and rerunnable loaders.
• Add deterministic reports for manager cycles, largest expense offenders, and supplier payment plans.
• Add isolated pytest coverage for the required expense report, source-data quality, manager-cycle output, and payment-plan behavior.
• Document the interpretations, rounding policy, expected results, run commands, and known limitations.

Key design decisions

• Treat the single exployee_id occurrence as a typo because the CSV and downstream requirements consistently use employee_id.
• Preserve the required invoice_ammount spelling.
• Omit receipt and invoice descriptions because the required destination schemas do not provide description columns.
• Calculate invoice due dates from Trino's current_date and run validation in UTC.
• Return one deterministic loop per actual cycle member and exclude employees whose manager chain only feeds into a cycle.
• Interpret an invoice due in N months as N month-end payments at offsets 0 through N - 1.
• Allocate residual cents to the earliest instalments, keeping payments within one cent and reconciling every invoice exactly.
• Aggregate supplier instalments by month before calculating balances with explicit window frames.

Testing and validation

• Trino: 483
• Image: trinodb/trino@sha256:db58cc93e593a2706553745f276bb119c9810e69918be56ecde088ba7ccb0534
• Executed all six submitted SQL files through Trino CLI standard input.
• Ran every loader twice to confirm rerunnable behavior.
• Ran python -m pip check: no broken requirements.
• Ran python -m pytest -q: 8 passed.
• Ran the state-mutating report tests individually and in both explicit orders.
• Confirmed mutation controls fail for an inclusive >= 1000.00 threshold and for a missing SQL file.
• Repeated the full SQL and pytest workflow from a fresh clone in a path containing spaces; Git status remained clean.

Performance and limitations

• Expense facts are aggregated before employee and manager joins.
• Supplier names are deduplicated before ROW_NUMBER() assigns IDs.
• Recursive traversal carries the next manager directly, requiring one hierarchy join per step; the supplied hierarchy is bounded to five steps.
• Payment instalments are grouped once before total and cumulative windows are evaluated.
• Dates remain intentionally relative to the Trino session date.
• The Memory connector is ephemeral, and the serial tests reset shared memory.default state.