Refactor get_next_set

[tomarrell]

"get_next_set's implementation is probably suboptimal.

You loop over the bitset until you reach a bit set to one. Granted the density of this bitset is high as per your choice of number of bits. Still it is difficult to predict, so that a lot of branch prediction error.

Instead you can use u64::trailing_zeros which compiles to a single instructions : https://github.com/tantivy-search/tantivy/blob/78673172d001a0c4c7c73cdd3d9923fc43fc0312/src/common/bitset.rs#L87" - fulmicoton