The CGI implementation starts to run into a problem:  Cache invalidation.
The state of all the cached filesystem stuff is going to get out of sync
with the underlying files.

Adds a couple of hashes in the App::WRT::EntryStore::new() constructor
which stash arrays of properties for each entry, and arrays of entries
for each property, as File::Find works its way through the $entry_dir.

Uses props_for() in App::WRT::entry_topic_list() instead of dir_list().
This is maybe a tiny performance gain, but also probably more hassle
than it's worth unless other features are going to be built on these
methods (or other code that calls dir_list() simplified).

This also moves a chunk of logic out of EntryStore::all() and into the
constructor, and directly accesses $self->{source_files} in a couple of
methods instead of tossing around copies of the array.

I'm kind of iffy on whether any of this is worth doing.

Fixes rendering of https://p1k3.com/2013/5/6/

I never could quite figure out why non-ASCII characters were getting
badly munged in XML::Atom::SimpleFeed output.  Using Encode::decode() on
the content of entries turns out to fix this, although it's probably the
tip of an iceberg.

- uses file_get_contents() for include processing
  - adds cache_includes setting
  - uses include_cache on wrt instance to stash file contents in memory

I didn't expect that this would be substantially faster than just letting
the kernel worry about it, but on a large set of files with includes in
the template, it shaves about 100ms off a render.

I'm not actually sure this is worth it.

Makes next() & previous() useful again.

- moves template existence check to new()
  - uses file_get_contents() instead of usual file slurping boilerplate
  - documents overlay option
  - deletes superfluous entry_wrapped()
  - tidies some commments

Already have a file_put_contents(); might as well.

- get_all_source_files() -> EntryStore::all()
  - get_date_entries_by_depth() -> EntryStore::dates_by_depth()
  - get_recent_entries_by_depth() -> EntryStore::recent_by_depth()

...plus some convenience methods for accessing recent or all years /
months / days, and a next($entry) / previous($entry) for indexing into the
date lists.

An instance of App::WRT::EntryStore is stashed on an App::WRT instance.

- give absolute paths to imgsize() so it chills out on Cwd::getcwd() calls
- rip out App::WRT::MethodSpit's convenience accessors and just access
  elements of $self directly
- kill local_path()
- stop using a() in entry_markup()
- cache get_date_entries_by_depth() results
- swap out state vars for stashing things on $self in
  get_all_source_files()

This gets rendering the whole p1k3 repo back down in the neighborhood of
2.1 seconds, which is a lot more tolerable than 5 or 6.

It also kinda technically breaks the API, maybe, so I guess this'll be a
v5.0.0 after a while.

All replaced with get_recent_entries_by_depth().

Replaces generate_month_lists with more generic generate_date_hashes(),
which now puts year, month, and day level entries in the prev/next hashes.

Strips previous links out of month() - these can live in templates now,
I guess, and work for year, month, and day.

Replaces get_all_day_entries() with get_date_entries_by_depth().  Adds
generate_month_lists() to stash %prev_months and %next_months (this can
probably be generalized a bunch).

Generalizes sortable_date_from_entry() to work with different levels of
date.  Assumes that no component of the date is more than 4 digits, which
is not very future proof.  (Though if this code is still running in the
year 10000, the future will have gone very differently than I expect it
to.)

month_before() now uses the %prev_months and %next_months hashes.

Removes fulltext(), because you can just call get_date_entries_by_depth(3).

Much slower than it probably needs to be.

Shuffles a bit of stuff out of App::WRT::Renderer into App::WRT, including
the method now called get_all_source_files() and a new one called
get_all_day_entries() that returns a sorted listed of days.

Uses state variables to cache the get_all_source_files() data, as well
as month_before() data (this was previously memoized inside a closure).

App::WRT::feed_print() now takes a list of entries instead of a month to
render the feed for.

App::WRT::feed_print_latest() gets a feed with most recent feed_length
entries.

It's weird to have this hardcoded, and there's
context on each entry anyway.

The default was `$0?`, which I think was probably a holdover from this
being invoked as a CGI script.  `$0` would give the name of the current
script, so you'd wind up with `display.cgi?2008` or what-have-you.