test: expand error path tests for float8

Add comprehensive tests for NaN propagation through all arithmetic ops
(Add, Sub, Mul, Div, Min, Max) and math functions, overflow clamping of
out-of-range float32 to infinity, underflow clamping to zero, and
zero sign handling across all four arithmetic operations including
division by zero and indeterminate forms (0*Inf, Inf-Inf).

Author: dndungu