daemon: initialize libjeffpc before printing the versions

This way they end up syslog'd as well.

Signed-off-by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>

include val.h instead of str.h

Signed-off-by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>

use libjeffpc's qstring parsing code

Signed-off-by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>

req: use const char * for cat/tag/feed/fmt members

There is no reason for these to be writable.

Signed-off-by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>

cmake: no need for a separate libversion

Since libblahg is a static lib, it can include the version info directly.

Signed-off-by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>

utilities & tests: no longer need an explicit libjeffpc initialization

Utilities and tests don't need the explicit initialization.  There were two
reasons blahg code did it:

(1) memory allocation purposes - making sure that the libjeffpc slab
    allocator was initialized
(2) setting up a log function - to direct errors to syslog

Recently, libjeffpc changed so that it uses shared object constructors,
therefore #1 is no longer a requirement.  And as far as #2 is concerned, the
utilities and tests don't want to spam syslog anyway.  Previously, they
accomplished this by setting BLAHG_DISABLE_SYSLOG=1.  Since not logging is
the default for libjeffpc, we can accomplish the same thing by not giving it
the custom ops vector and relying on the defaults.

Signed-off-by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>

index: don't print the archive timestamp bounds

This was a debug message that was useful when developing the function, but
it is pure noise at this point.

Signed-off-by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>

Good news, everyone!

The sqlite3 database that was causing all sorts of cache invalidation
related headaches is gone!  Instead, the daemon just reads everything into
memory.  This seems to be marginally faster and it seems to scale better
too.  Only time will tell if this was a good idea.

MIT license

After way too much thought, I decided that the MIT license is the way to go.
This is the first version that has any license on it.  Quite exciting!

Other than that, the one big (technical) change is the logging of every
request.  I expect this to get useless/tiresome, but until then...disk is
cheap.

Conditional templates

Yep, after years of living without conditional markup in the template
language, I finally bit the bullet and implemented simple conditional logic.
It doesn't support much, but it supports enough to get rid of the two
biggest annoyances: the "previous page" and "next page" links going off to
infinity, and posts with no comments failing to validate because of an empty
<ol> tag.

Other than than, a couple of memory leaks were addressed, the str and req
structures are now allocated using umem caches, and finally the requests
keep track of assorted latencies.