Remove and replace misleading Time.wall_to_nanos

[SeanTAllen]

Based on its description and name, one would be reasonable in expecting
Time.wall_to_nanos to take a wall clock time and convert it the number
of nanoseconds since the "start of time". That isn't however what it
does. It was getting an absolute time to fire a timer setup in
Timer.abs.

This PR correct this problem by moving the functionality that previously
existed in wall_to_nanos into Timer as a private method and adds a
function to Nanos to convert from a wall clock time to nanoseconds.

[SeanTAllen]

I believe this breaking change should count as a principle of least surprise bug fix.

[jemc]

Your reasoning sounds right to me as well, but I'd like some other(s) to agree before merging, since it is a breaking change, and I'm not totally confident that I'm not missing something.

[SeanTAllen]

Serialization tests for this failed across the board. I believe its the work that @Praetonus has been doing. Its not clear to me why they failed.

[SeanTAllen]

In particular: CodegenTest.CustomSerialization fails.

It fails for me locally a well but master does not. I'm totally stumped why that would happen.

[Praetonus]

The error is

Error:
/home/travis/build/ponylang/ponyc/packages/time/timer.pony:59:39: receiver type is not a subtype of target type
    _expiration = _abs_expiration_time(expiration)
                                      ^
    Info:
    /home/travis/build/ponylang/ponyc/packages/time/timer.pony:59:19: receiver type: Timer tag
        _expiration = _abs_expiration_time(expiration)
                      ^
    /home/travis/build/ponylang/ponyc/packages/time/timer.pony:117:3: target type: Timer box
      fun _abs_expiration_time(wall: (I64, I64)): U64 =>
      ^
    /home/travis/build/ponylang/ponyc/packages/time/timer.pony:59:19: Timer tag is not a subtype of Timer box: tag is not a subcap of box
        _expiration = _abs_expiration_time(expiration)

It looks like not every field of Timer is initialised when calling _abs_expiration_time in a constructor, resulting in the object having a capability of tag.

I think we should add a more detailed error for those cases, something like this would be possible if all fields were intialised before calling the function.

[SeanTAllen]

@Praetonus why does that cause custom serialization to fail?

[SeanTAllen]

Pushing fix.

[jemc]

I agree with @Praetonus that we can do better with that error message.

It should be relatively to provide that extra information, because since the refer pass was added, we are setting the AST_FLAG_INCOMPLETE flag on the TK_THIS node to indicate that not all fields are initialized yet.