class Float has 3 constructors. One takes a String argument and parses it. The second comstructor takes a float and the third takes a double.
In converting an integer type, the "Float(float value)" constructor will be invoked because it has the most restrictive parameter type that still accepts an integer type. float is more restrictive than double because a float argument will be accepted as a double parameter without an explicit cast, while a double argument will not be accepted as a float parameter without an explicit cast.
The concept of a widening cast has to do with the range of legal values, not precision. So any integer type can become a float as a widening conversion, no explicit cast required. It's true that a double would preserve more precision from an int or long, but Java only considers range.