[cdc] Reject a non-positive width for the truncate computed column - #10360

Purpose

The truncate computed column (Expression.TruncateComputer) parses its
width literal but never checks the sign. A non-positive width is only caught
later, at per-row eval, and inconsistently: integer and DECIMAL columns
throw ArithmeticException: / by zero once the sync job is already running,
while CHAR/VARCHAR columns silently produce an empty string. A width of 0
or a negative value is never meaningful for truncation.

This validates width > 0 in the constructor, so a bad definition such as
--computed_column 'x=truncate(amount,0)' fails fast at job build time with a
clear message, mirroring the sibling substring expression that already
validates its indices.

Tests

TruncateComputerTest#testTruncateRejectsNonPositiveWidth asserts the
constructor rejects width 0 and negatives across int/bigint/decimal/varchar.
Existing positive-width coverage is unaffected.

API and Format

No change.

Documentation

No change.