On the Mac, we have to ship these along with libquicktime itself, and
define LIBQUICKTIME_PLUGIN_DIR in the environment so that libquicktime
finds them. Otherwise Gem's QT4L plugins won't be of much use.

This requires various changes throughout the build system. Moreover,
the Mac build now works with both Homebrew (in /usr/local/opt) and
MacPorts (/opt/local), but still prefers Homebrew if it is present.

We also adjusted the embed-MacOSX-dependencies.sh script to look for
libquicktime dependents in the right place ($PD_APP_CONTENTS/Plugins),
so that Gem's QT4 plugins will work using the Homebrew libquicktime
package. Alas, MacPorts doesn't have a working libquicktime package, so
the QT4 plugins are disabled there.

These are now proper errors instead of warnings, and the error message
on Windows is more informative as well, making it easier to detect and
diagnose such errors.

Backport from upstream:

https://github.com/pure-data/pure-data/commit/fee700b3
https://github.com/pure-data/pure-data/commit/af3894da

This adds all generated markdown and html docs for all target systems in
packages/gendoc to the self-contained distribution tarball (`make
dist`), so that no special tools like python-markdown are required to
build the package.

Built-in flag and live usage data are now combined into a single numeric
relevance score, with a boost factor determining the weight of the
former relative to the latter. This makes it easy to adjust the
relevance order. Currently we have the boost factor at 50, which means
that an external needs to have 51 uses before it gets preferred over a
built-in with no uses at all. Is this too large? Too small? Also, should
the fuse similarity score take precedence over relevance (currently it
doesn't)? At present nobody knows, so let's give it a try and listen to
user feedback to improve it.

If we find any LIBRARY meta data, record it in the completion index.
Currently this is only used to list built-ins ("internal" library) first
if sorting by relevance is enabled.

We now properly deal with argument completions scraped from the NAME
field, by checking for existing objects and argument completions, and
updating the object accordingly.

If enabled, this orders object and argument completions first by
relevance, then alphabetically by object name, with "relevance"
currently being defined as frequency of use (as given by the occurrence
field in the completion data).

As usual, the option is stored in the user preferences, its checkbox can
be found on the GUI Preferences tab.

This effectively gives a way to scroll the popup using Tab to cycle
through the completions, without having to add a scrollbar to the menu.

While we were at it, we also added some new keybindings for Alt+Home and
Alt+End which quickly take you to the first and last completion,
respectively. It's easy to get lost in the completion list if it gets
very long (try completing `list` some time), so this makes navigating
the list much easier.

The completion popup only shows the first few completion items, because
of limited screen real estate. But there's no reason that Tab completion
should have the same limitation. So we now keep track of the full list
of completions internally and let Tab cycle through them all.

Fuse's result lists are ordered according to similarity scores, which
may look pretty arbitrary at times. Since we only display at most 8
results at a time, this is a problem, because a rather random selection
of completions may be shown in the menu. To improve this, we now also
order results for the same score alphabetically, which makes the results
look much more sensible IMHO.

As soon as the user types a blank after the object name, we now
immediately show the list of available argument completions. This makes
it easier to discover an argument completion when you're not sure how it
begins.

Deal with a bunch of special cases which have multiple objects
documented in them. In this case we extract the object names from the
NAME field of the META data and enter them into the completion index.

As generation of the autocompletion index is tied in with the generation
of the help index, building the index can't be deferred until the help
browser is opened if autocompletion is enabled. Previously this resulted
in missing completions or autocompletion being completely unavailable
until the help index was (re-)built. This commit addresses this bug by
ensuring that the help index is generated right away when needed.

There are reasons why you may want to get rid of completions. Sometimes
the autocompletion index may contain completions that are not actually
valid object names, or arguments that you may no longer wish to use.
This commit introduces a new key binding for ctrl+y which lets you
remove a completion from the index. (Note that ctrl+y is also used for
the Tidy Up operation in the Edit menu, but the two bindings don't
interfere with each other.)

If a valid completion has been entered, ctrl+y will ask in the Pd
console to confirm that you really want to remove the completion.
Pressing ctrl+y again will remove the completion and report this in the
console. Pressing Esc or any other key will cancel the operation.

Note that we keep this low-profile by design. While a GUI implementation
would be possible, the keyboard + console interface is quick and
unobtrusive, and also much easier to implement.