Fix numeric to float64 conversion

[jschaf]

A previous optimization in 50933c0 had a logic bug when converting pgtype.Numeric to a float64:

        // BUG: An exponent of one means multiply by 10.
	if src.Exp == 1 {
		return float64(src.Int.Int64()), nil
	}

I think the current pgtype.Numeric float64 conversion is incorrect for integers with an exponent of 1, that are divisible by 10, meaning [10, 20, 30, ..., 90].

#210

[jschaf]

This optimization isn't present in pgx@v5 so I don't think there's anything to fix in v5.

https://github.com/jackc/pgx/blob/20bf953a17fc7cf6619536a87bd285b9273dbcf9/pgtype/numeric.go#L66

[jschaf]

Please double-check my logic. I think src.Int is normalized, so this optimization is valid.

[jackc]

I think there is a danger of integer overflow in float64(src.Int.Int64() * 10). I think it should be float64(src.Int.Int64()) * 10 instead. But more so, is that branch even necessary / valuable. I assume it would work without it.

[jschaf]

I removed the case for exp == 1 since you're right; it's not necessary. I originally kept that case since I thought it was important for the original optimization.

If you wanted to get fancy, you could implement it for all positive exponents, but I'll leave just the case for exp == 0 for now.

	if src.Exp >= 0 {
		return float64(src.Int.Int64()) * math.Pow10(int(src.Exp)), nil
	}