Some remaining cases of fractional inputs being converted silently to integer dimensions:

filter (ones(3,3,3,3), 2, ones(3,3,3,3), [], pi)   # Last argument is dimension

cellfun ("size", cell(3,3), pi)                    # Last argument is dimension

A = rand (3, 3); D = eigs (@(x) A*x, pi, 1)        # Penultimate argument is dimension

rats (e, 12.345)                                   # Last argument is string length

Changing `nint_value` to `strict_int_value` causes these functions to give errors for those fractional inputs and also continue to trap nan inputs like they did with nint_value, which overall seems better.

I'll push that change once I get a bug number.

Inputs of +inf and -inf are still allowed by strict_int_value, exactly like nint_value. Whether that constitutes legal input depends on the individual function. Some functions have additional checks after the call to integer conversion to check for inf.

Note that this code:

rats (pi, N)

asks Octave to construct a rational approximation to `pi` whose string representation is N characters long. Setting N to a very large value or inf causes an OOM crash, also on stable. Not sure whether to guard against it, since such code specifically asks for it...