Fix overflow during indexing for data with more than 2B entries

[wlattner]

This PR fixes an overflow during indexing. This causes the error below when reading a file with more than 2B entries.

m <- data.frame(x1 = rep(0, 1e9), x2 = rep(1, 1e9), x3 = rep(2, 1e9))

f <- tempfile()
vroom::vroom_write(m, f)

df <- vroom::vroom(f)
#> Error in vroom_(file, delim = delim, col_names = col_names, col_types = col_types,  : 
    Failure to retrieve index 184,467,440,724,145,845 / 6,148,914,690,804,861,440 

I am able to reproduce this on MacOS with clang 6 and Ubuntu with gcc 7.4. In both cases, the literal 0 starting value for std::accumulate is interpreted as int rather than size_t.

[jimhester]

Thanks! Looks good!
Great work, this package is now better thanks to you!