coerce numbers when appropriate

[philhofer]

Addresses some comments in #134
@client9

[klauspost]

@philhofer I got a bit nerd-sniped by this.

It seems like we can also return float values that map exactly to an integer - meaning a value that is exactly the integer without any rounding.

// isExactInt will return true if the number represents an integer value.
// NaN, Inf returns false.
func (n *Number) isExactInt() bool {
	var eBits int // Exponent bits
	var mBits int // Mantissa bits

	switch n.typ {
	case InvalidType, IntType, UintType:
		return true
	case Float32Type:
		eBits = 8
		mBits = 23
	case Float64Type:
		eBits = 11
		mBits = 52
	default:
		return false
	}
	// Calculate float parts
	exp := int(n.bits>>mBits) & ((1 << eBits) - 1)
	mant := n.bits & ((1 << mBits) - 1)
	if exp == 0 && mant == 0 {
		// Handle zero value.
		return true
	}

	exp -= (1 << (eBits - 1)) - 1
	if exp < 0 || exp == 1<<(eBits-1) {
		// Negative exponent is never integer (except zero handled above)
		// Handles NaN (exp all 1s)
		return false
	}

	if exp >= mBits {
		// If we have more exponent than mantissa bits it is always an integer.
		return true
	}
	// Check if all bits below the exponent are zero.
	return bits.TrailingZeros64(mant) >= mBits-exp
}

It doesn't check if the value will under/overflow a given integer size, though.