Support microsecond precision for time stamps.

[paurkedal]

According to the MySQL 8.0 documentation, second_part field represents microseconds, so presumably this is set in stone, but if you prefer, we could keep the second_part name in OCaml, use int64 instead of int and expose SEC_PART_DIGITS.

I had to put ?microsecond before ~second to make it truly optional without fully breaking backwards compatibility. Alternatively we could add _us variants to to all functions taking ~second arguments.

[andrenth]

I think the way you implemented it in the PR is great. Thanks for the feature!

[paurkedal]

I realise only now that this PR breaks backwards compatibility. The issue is the the a mandatory named argument does not cancel previous optional arguments like a unnamed argument would do,

utop # let f ?(a = 0) ~b = a + b;;
val f : ?a:int -> b:int -> int = <fun>
utop # f ~b:3 = 2;;
Error: This expression has type int but an expression was expected of type
         ?a:int -> int

That means making the argument optional also does not serve much purpose, as one might as well write ~microsecond:0 as ?microsecond:None. I am sorry for the oversight.

The question then is how to resolve it, assuming you agree that the current signature is not optimal. The most common practise, as far as I have seen, is to add at least one unnamed argument, even if it is unit. My suggestion would be the straight forward adjustment

val datetime : year: int -> month: int -> day: int -> hour: int -> minute: int -> second: int -> ?microsecond: int -> unit -> t

Another option would be to make the microsecond mandatory. I can send a new PR, but I'd like to know you preference first, and also whether the date and time should be adjusted for consistency.

[andrenth]

Petter, I think the extra parameter is the way to go, along with keeping everything consistent in date and time.

[paurkedal]

Good I created #23.