Add tolerant double comparison

Allow optional non-negative relative and absolute tolerances for reporting floating-point cell changes.

For finite x and y, compare equal when:

|x - y| <= max(t_abs, t_rel * max(|x|, |y|))

Requirements:

• Both tolerances default to zero.
• Tolerance applies only to cell-level value-change reporting.
• Key matching, hashing, and rename inference remain exact because approximate equality is not transitive.
• The UI displays active tolerances so omitted small differences are explainable.

Start by tracing the cell-level value-change reporting path and the UI that presents differences. Implement zero-default non-negative absolute and relative tolerances only for value changes, while keeping key matching, hashing, and rename inference exact; done means active tolerances are visible in the UI and the stated finite-value comparison rule is covered.