s Aho-Corasick
generally maintains reasonably consistent performance regardless of the number
of patterns you give it. Because of this, the vectorized searcher in this
sub-module cannot be used as a general purpose searcher, since building the
searcher may fail even when given a small number of patterns. However, in
exchange, when searching for a small number of patterns, searching can be quite
a bit faster than Aho-Corasick (sometimes by an order of magnitude).

The key take away here is that constructing a searcher from a list of patterns
is a fallible operation with no clear rules for when it will fail. While the
precise conditions under which building a searcher can fail is specifically an
implementation detail, here are some common reasons:

* Too many patterns were given. Typically, the limit is on the order of 100 or
  so, but this limit may fluctuate based on available CPU features.
* The available packed algorithms require CPU features that aren't available.
  For example, currently, this crate only provides packed algorithms for
  `x86_64` and `aarch64`. Therefore, constructing a packed searcher on any
  other target will always fail.
* Zero patterns were given, or one of the patterns given was empty. Packed
  searchers require at least one pattern and that all patterns are non-empty.
* Something else about the nature of the patterns (typically based on
  heuristics) suggests that a packed searcher would perform very poorly, so
  no searcher is built.