Small doubles are serialized incorrectly by fwrite

[st-pasha]

> # This is correct, up to some unavoidable loss in precision
> data.table(A=1e-323)
               A
1: 9.881313e-324
> # But this is not
> fwrite(data.table(A=1e-323))
A
1.1125369292536e-308

[mattdowle]

> .Machine
$double.xmin
[1] 2.225074e-308

So I wouldn't say this is particularly urgent for next release, since 1e-323 < .Machine$double.xmin. Moving to candidate milestone.