!folder name was not working because strlen("name") was expected to be
5. Expect it to be 4 instead.

This moves the folder index, text, moderation and tmp files into
directories corresponding to the file name suffixes currently in use.

Old                     New
---                     ---
STATEDIR/foo.i          STATEDIR/index/foo
STATEDIR/foo.i.m        STATEDIR/index/mod/foo
STATEDIR/foo.i.t        STATEDIR/index/new/foo
STATEDIR/foo.t          STATEDIR/text/foo
STATEDIR/foo.t.m        STATEDIR/text/mod/foo
STATEDIR/foo.t.t        STATEDIR/text/new/foo

This has the following advantages:
1. No more ugly folder path building all over the place
2. STATEDIR and PATHSIZE used in far fewer places
3. No longer assumes paths to folder files will fit in 256 chars
4. A tidier STATEDIR
5. Better separation of concerns (more folder stuff hidden in folders.c)
6. Should make it easier to move the folder stuff into the server
7. Improved error checking/reporting

This makes use of the *at family of syscalls so, instead of building
paths, we now use file descriptors.

An earlier patch accidentally left a zero userposn in ipc_connect but we
don't have a userposn now so ipc_connect now accepts a struct user to
identify as. This makes reconnecting in ipc_check() a little trickier so
pass the stored target in from outside instead of checking for NULL.
Also add some error checking.

Also switch to pread.

We're using user->posn now.

Wrap the userposn and struct person in a new user structure which can be
used to pass around all of the date related to the user, not just the
on-disk bits. This should make things easier down the line when
replacing the users.bb format with a more flexible one.

The control file needed to include more build deps on a ssytem with a base netinstall and build-essential.

"There is a limit to the time assigned you, and if you don't use it to
free yourself it will be gone and never return."
                                               -- Marcus Aurelius

Unreturnable gag

See merge request !3

Ignore the database being busy.

This fixes issue https://projects.sucs.org/arthur/mw/issues/3 by just ignoring
busy messages when the DB get overloaded by spamming messages with a hashtag.

See merge request !1

Not checking against a variable.

This fixes issue https://projects.sucs.org/arthur/mw/issues/3 by just ignoring
busy messages when the DB get overloaded by spamming messages with a hashtag.

Now the client can use it too if necessary.

Move the bits which specified the default server options and config file
paths out of the config system. Add some bits to simplify constructing an
array of default config options.

Dedupe it into src/user.c where it fits and rewrite it in terms of a new
user_find_name function which will be used more later.

Move it from src/{web,}client/user.[ch] into src/user.c and simplify
with pwrite(2).

Until now configuring mwserv has been limited to command line options
and build configuration. This config system should make things a bit
more flexible by adding the ability to load bits and pieces from a
json config file (or many json config files).

This is how it works:
1. A json object 'cfg' is built containing the build-time defaults for
   each config option recognised by mwserv. This will be used to
   validate the options in config files later.
2. A json config file in the user's home directory is loaded if it
   exists. It does not have to be complete: its contents will be merged
   into the defaults. Unrecognised options are ignored.
3. Otherwise, a config file in /etc is loaded and used similarly.
4. Command line options are parsed and the cfg object is updated with
   cfg_set_* in the order in which the options are found.
   4a. If the [--config|-c <file>] option is encountered, the file is
   loaded and merged into the cfg object as before.
5. Once all of the config information is gathered and merged into the
   cfg object, if the [--print-config|-P] option was given, the final
   merged config is printed in json format (mwserv then exits).
6. Config options are then accessed using the cfg_get_* type-specific
   functions.

This should allow the build system to be simplified in future, e.g.
'make test' could go away in favour of using ~/.mwserv.conf to store
the test environment configuration.

For now, the place to add new config options and their default values is
in cfg_init_defaults() in servercfg.c.

Tip: this can be avoided by adding this to your .vimrc:

match ErrorMsg '\s\+$'

Adding tags to the root .gitignore will make git ignore tags files in
any subdirectory so there's no need for a second .gitignore.