Add RowsColumnScanner support (Go 1.27+)

[Copilot]

Implements the driver.RowsColumnScanner interface introduced in Go 1.27, which allows drivers to scan row data directly into user-provided destinations, bypassing the intermediate driver.Value allocation path.

Changes

• rows.go: Reverted to original (no build tag). Adds rawCols [][]byte to mysqlRows for use by the RowsColumnScanner implementation. Unused struct fields are permitted by Go, so no build constraint is needed.
• rows_column_scanner.go (//go:build go1.27): Implements the interface for both row types without calling readRow():
  • (*textRows).NextRow() — reads the packet and stores raw length-encoded byte slices (subslices of the packet buffer) into rawCols, with no intermediate []driver.Value allocation.
  • (*binaryRows).NextRow() — reads the packet, processes the null bitmap, and extracts per-column byte subslices based on the binary wire format into rawCols.
  • (*textRows).ScanColumn(scanCtx, i, dest) — fast paths for *[]byte/*sql.RawBytes, then parses bytes by fieldType and delegates to sql.ConvertAssign.
  • (*binaryRows).ScanColumn(scanCtx, i, dest) — parses binary wire bytes per fieldType (little-endian integers, IEEE-754 floats, length-encoded strings, binary date/time) and calls sql.ConvertAssign directly.

The key allocation benefit: rawCols is allocated once as [][]byte and reused across rows, while each element is a zero-copy subslice of the packet buffer. No []driver.Value is allocated per row. sql.RawBytes destinations receive zero-copy access to the packet buffer.

When Go 1.27+ is used, database/sql automatically prefers NextRow + ScanColumn over Next([]driver.Value). For older Go versions, behavior is unchanged.

[coveralls]

coverage: 82.731% (-0.07%) from 82.804% — copilot/support-rows-column-scanner into master