fix: restore integer types in psycopg/psycopg2 replay using column metadata

[jy-tan]

Summary

Fixes integer-to-float conversion that occurs during the record/replay cycle due to JSON serialization. When database rows are recorded, they pass through JSON serialization where Go's json.Unmarshal converts all numbers to float64. During replay, this causes integer values like 0 to be rendered as 0.0 in HTML responses, causing test deviations.

This fix uses the PostgreSQL column type_code (OID) stored in the cursor description to safely identify INTEGER columns and restore their types during replay.

Changes

• Add POSTGRES_INTEGER_TYPE_CODES constant with PostgreSQL OIDs for integer types (SMALLINT, INTEGER, BIGINT, OID, XID)
• Add restore_row_integer_types() function that uses column metadata to convert whole-number floats back to int for INTEGER columns only
• Update psycopg2 instrumentation to restore integer types after deserializing mock rows
• Update psycopg3 instrumentation to restore integer types in both _mock_execute_with_data and _mock_executemany_returning_with_data

Technical Details

The type restoration is safe because:

• It only converts values that are whole-number floats (value.is_integer())
• It only converts columns with INTEGER type codes from PostgreSQL
• FLOAT/DOUBLE PRECISION columns (type codes 700, 701) are never converted
• Column metadata (description) is already recorded and available during replay

[cubic-dev-ai]

No issues found across 3 files