store distances as double rather than float

[corybrunson]

This PR changes the value_t type and ratio parameter from float to double.

Benchmark comparisons on small and medium data sets incurred no consistent cost to storage or runtime.

Coordinated with this ripserq PR.

[corybrunson]

@StrongKs @SeanHersh @resetdisconnect could you check that this branch installs and works on your machines? Feel free to experiment with the benchmark script but be cautious with the larger o3 data sets.

[corybrunson]

This change requires a unit test, and in designing one i observed something strange. With v1.0.0 installed, the following code detects differences only for digits < 8; whereas, with the double branch loaded, it detects differences for all values of digits as well as with just d <- dist(a)—though, notably, with digits = 14 or = 15, only the second death time is detectably different.

# equilateral triangle
a <- rbind(
  c(0,0),
  c(sqrt(3)/2,.5),
  c(sqrt(3)/2,-.5)
)
d <- dist(round(a, digits = 15))
p <- as.data.frame(vietoris_rips(d))
p$death[1:2] == 1
p$death[1:2] - 1