Add a hash table to speed up bee_user_by_handle()
This maintains a hash table next to the linked list, which results in
negligible additional memory usage (~300kb for 10k users) but allows
instant lookups.
This was a big problem with discord, which has huge user lists and joins
everyone to every channel. In my test, the GUILD_SYNC event for 10k-50k
user lists is now approximately 5 times faster.
This hash table based code is only used if handle_cmp is either
exact or case-insensitive string comparison (g_ascii_strcasecmp or
strcmp/g_strcmp0).
The old function that goes through the bee->users linked list is now
called bee_user_by_handle_slow() and used for protocols with unusual
handle_cmp functions - skimming through the code, just oscar.
May revisit this if it happens to more meaningful protocols.
The case-insensitive hashtable functions are copied from irssi, which is
also GPLv2. I renamed them from g_ to b_ (g_istr_equal to b_istr_equal)
