The number of separate buckets for weights (discretization). The more buckets,
the more fine-grained term weights (priorities) can be assigned. The speed of lookup
will not decrease for prefixes which have highly-weighted completions (because these
are filled-in first), but will decrease significantly for low-weighted terms (but
these should be infrequent, so it is all right).
The number of buckets must be within [1, 255] range.