Consider adding `default` property to altrep class

[EmilHvitfeldt]

right now there is a implicit assumption that non-specified values are zero. This doesn't have to be true, and we could include default as a value to data1 and use. With this we can have x + 1 retain its sparsity which is currently not the case.

library(sparsevctrs)

x <- new_sparse_real(c(1, 5, 10), c(1, 5, 10), 10)

x
#>  [1]  1  0  0  0  5  0  0  0  0 10

altrepr::is_altrep(x)
#> [1] TRUE

x + 1
#>  [1]  2  1  1  1  6  1  1  1  1 11

altrepr::is_altrep(x + 1)
#> [1] FALSE

We could also take variables that mostly take 100 as a value to become sparse with default 100

[EmilHvitfeldt]

To make this easier, I'm going to make it so all the coerce_ functions stop if the sparse vectors doesn't have default value 0