fix %in% and negative 'nanoperiod' parse. closes #96 #99

[lsilvest]

Changing the type on which nanotime is based from integer64 to numeric does not solve the dispatch problems. It looks like there is a more fundamental dispatch issue in R when S3 methods are defined outside of base.

Here I have solved the issue by providing both S3 and S4 methods for %in%. When bit64 is not explicitly loaded and no S3 method exists, the S4 method is invoked. After bit64 is loaded the S3 method is invoked because if it is not present the bit64 S3 method will trump the nanotime S4 method.

Also fixed the parsing of a nanoperiod when it consists of only its negative nanoduration part.

[eddelbuettel]

Changing the type on which nanotime is based from integer64 to numeric does not solve the dispatch problems

Ack. There is clearly some voodoo there. I also see some packages do S3 registration tricks at load time by I am not all that fluent wuth that trick either.

But getting rid of the 'noise' at load time is ❤️

[eddelbuettel]

LGTM. Wild how the entire change is (essentially) the method declaration (and reactivation of the parked code).

[eddelbuettel]

I fixed a micro-typo/mis-edit in NEWS.Rd and am updating the docs. I think we can ship this to CRAN as is otherwise.