Simplify/speedup "--show" and "--make-charset"

As I pointed out in https://github.com/openwall/john/issues/4388#issuecomment-719723533, jumbo's usage of split() on --show is much heavier than core's, which is both good (allowing for #4428) and bad (slow). #4429 is a start at optimizing it, however we can do more.

Perhaps most importantly, we can optimize the scanning of the formats list in ldr_show_pot_line() to cache and probe the previous line's format's valid() first.

Also, I think right after merging #4429 (or adding to it) we can also make ldr_cracked_hash() case-sensitive.

Further, why do we(?) even bother to build the db->cracked_hash table when invoked with --make-charset? We should probably skip that.

Finally, given that #4429 obsoletes FMT_SPLIT_UNIFIES_CASE, should we drop that flag from everywhere, keep it around in case we need it again later, or rename/repurpose it (e.g., to just FMT_SPLIT_UNIFIES, so that we'd set it in more formats yet be able to skip split() in ldr_show_pot_line() for most formats)?

Start by reading ldr_show_pot_line(), ldr_cracked_hash(), the --make-charset path, and uses of FMT_SPLIT_UNIFIES_CASE. Compare the behavior and cost of --show and --make-charset before deciding which optimizations and flag changes are still appropriate; done means the selected changes preserve existing behavior while reducing unnecessary work.