Returning tensors lists much faster.

[dfalbel]

Fixes #992

> bench::mark(a = {splitted <- torch::torch_split(sortedTensor[,2], counts)})
# A tibble: 1 × 13
  expression      min   median `itr/sec` mem_alloc `gc/sec` n_itr  n_gc total_time result          memory     time           gc      
  <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl> <int> <dbl>   <bch:tm> <list>          <list>     <list>         <list>  
1 a             177ms    177ms      5.58    1.15MB        0     3     0      538ms <list [74,999]> <Rprofmem> <bench_tm [3]> <tibble>

compared to

# A tibble: 1 × 13
  expression      min   median `itr/sec` mem_alloc `gc/sec` n_itr  n_gc total_time result          memory     time           gc      
  <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl> <int> <dbl>   <bch:tm> <list>          <list>     <list>         <list>  
1 a             1.11m    1.11m    0.0150      21GB     4.12     1   275      1.11m <list [74,999]> <Rprofmem> <bench_tm [1]> <tibble>