Option to only read in double numerics

[mmuurr]

I think it would be useful to have a flag that would bypass the "does the value look like an integer" logic, opting instead to always try to create doubles during parsing. If a value larger than 2^53 happens to be in the data, this should probably be converted to Inf (or -Inf), though NaN seems fine, too ... as since this would be an opt-in flag the onus is on the programmer to know their data is representable by a float64.

The reasoning here is just to help guarantee type-consistency when parsing, e.g., list-cols in a data frame, where some entries in the source list-col just happen to contain vectors of integerish values, while other entries have non-integerish doubles.